Search

Physical paper, in its various forms (e.g. books, leaflets, catalogues), is extensively used in everyday activities, despite any advancements in digital technology and ICT. Inspired by the popularity of this medium, several research efforts have envisioned and pursued a new era of interactive paper, however several challenges still remain to be addressed. On the other hand, recent advancements towards Ambient Intelligence (AmI) and Smart environments, bear the promise of seamless integration of the physical and digital worlds in an intuitive and user-friendly manner. This paper presents InPrinted, a systematic and generic framework supporting physical paper augmentation and user interaction in Ambient intelligence environments. InPrinted enables natural multimodal user interaction with any kind of printed matter in smart environments, providing context aware and anticipation mechanisms, as well as tools and interaction techniques that support the development of applications incorporating printed matter augmentation. InPrinted has been put to practice in the development of various systems, including the Interactive Maps system, facilitating interaction with printed maps as well as their augmentation with digital information. The results of an in-situ observation experiment of the Interactive Maps system are reported, highlighting that interacting with augmented paper is quite easy and natural, while the overall User Experience is positive.

Paper checks may have complex background features including fine lines and patterns, which make forgery more difficult. Also, halftoning techniques are used to produce continuous tones and to prevent copies with void pantograph features. When these kinds of checks are scanned, Moire patterns may occur. These patterns make it difficult for customers to examine the scanned check images on ATM (Automated Teller Machine) displays. They also can decrease the classification accuracy of check recognition systems. In this paper, we propose an algorithm to enhance the perceptual quality of scanned check images by reducing the Moire patterns. The proposed algorithm consists of foreground extraction, Moire detection and Moire removal. Subjective image quality assessment was performed to evaluate the degree of improvement. Experimental results show that the proposed algorithm improves perceptual quality while maintaining check recognition accuracy.

International audience; This paper presents a novel method for image manipulation iden- tification of natural images in JPEG format. The image forgery detection technique is based on a signal-dependent noise model that is relevant to de- scribe a natural image acquired by a digital camera. This parametric model is characterized by two fingerprints which are used to falsification identification. The problem is cast in the framework of the hypothesis testing theory. For practical use, the Generalized Likelihood Ratio Tests (GLRT) are presented and their performance is theoretically established. There are different types of image forgery which have been considered in this paper for example re- sampling, Gaussian filtering and median filtering. Experiments with real and simulated images highlight the relevance of the proposed approach.

This paper considers the area of digital forensics (DF). One of the problem in DF is the issue of identification of digital cameras based on images. This aspect has been attractive in recent years due to popularity of social media platforms like Facebook, Twitter etc., where lots of photographs are shared. Although many algorithms and methods for digital camera identification have been proposed, there is lack of research about their robustness. Therefore, in this paper the robustness of digital camera identification with the use of convolutional neural network is discussed. It is assumed that images may be of poor quality, for example, degraded by Poisson noise, Gaussian blur, random noise or removing pixels’ least significant bit. Experimental evaluation conducted on two large image datasets (including Dresden Image Database) confirms usefulness of proposed method, where noised images are recognized with almost the same high accuracy as normal images.

Prostate cancer (PCa) has become the second most dreadful cancer in men after lung cancer. Traditional approaches used for treatment of PCa were manual, time consuming and prone to subjective errors. Thus, there is a need for a Computer aided diagnosis system (CADs) consisting of denoising, segmentation, and classification approaches for diagnosis of PCa. CADs may act as a second opinion for the medical experts and save their precious time used in manual analysis. Magnetic resonance imaging (MRI) is the commonly used modality, as it produces detailed and fine contrast images of internal organs for diagnosis of PCa, but it may contain a certain amount of rician and gaussian noise which is necessary to be denoising before segmentation and classification. Denoising offers several challenges such as suppressing of significance image details leading in inaccurate segmentation and classification for prediction of abnormality. Thus, improved denoising, segmentation, and classification approaches can overcome the challenges by analyzing the pitfalls in the state of the art. This paper presents the experimental analysis state of the art denoising and segmentation approaches to analyse their performance based on the values of Peak signal to noise ratio (PSNR), Mean squared error (MSE), Structured similarity index (SSIM), dice metric, area overlap and accuracy. Based on the experimental analysis it was analysed that anisotropic filter outperforms other filters for gaussian noise with PSNR of 28.29, MSE of 96.22 and SSIM of 0.64. Also, for the rician noise anisotropic filter outperforms others with PSNR of 28.06, MSE of 101.52 and SSIM of 0.01. Similarly, for the combined gaussian and rician noise, anisotropic filter outperform others with PSNR of 28.34, MSE of 95.13 and SSIM of 0.652. Further, the analysis of segmentation approaches such as contour and shape-based, region/atlas based, thresholding based, clustering based and deep learning based was performed. Amongst these approaches deep learning based segmentation was found to outperform with dice metric of 0.89 and area overlap of 0.80. Also, CNN based classification outperformed machine learning based Support vector machine (SVM), K nearest neighbour (K-NN) and Random forest (RF) with 94.55% sensitivity, 93.34% specificity, 95.45% accuracy. Finally, the paper discusses challenges and future scope based on analysis in the concerned field for diagnosis of PCa.

Detecting and classifying a brain tumor is a challenge that consumes a radiologist’s time and effort while requiring professional expertise. To resolve this, deep learning techniques can be used to help automate the process. The aim of this paper is to enhance the accuracy of brain tumor classification using a new layered architecture of deep neural networks rather than the current state-of-the-art algorithms. In this paper, we propose automated tumor classification by concatenating two convolutional neural network structures of layers and tuning the hyperparameters by utilizing Bayesian optimization. The proposed solution focuses on enhancing the accuracy of classifying tumors to increase the level of trust in the technologies employed in the medical field. The work is tested and evaluated to predict the classification of magnetic resonance imaging inputs and achieving a higher accuracy (97.37%) than other similar works, with accuracies between 84.19% and 96.13%, for the same dataset.

The goal of learning-based image super-resolution (SR) is to generate a plausible and visually high-resolution (HR) image from a single low-resolution (LR) input image. The SR is an important branch of image reconstruction that concentrates on the improvement of image resolution and the problem is severely underconstrained. It’s to rely on examples or some strong image priors to reconstruct the missing HR image details. The paper addresses the problem of learning the mapping functions is, projection matrix between the LR and HR images based on a dictionary of LR and HR exemplars. In this paper, we present a self-learning single image SR method, which restores an HR image from self-examples extracted from the LR input image itself without relying on additional external training images. This paper proposes a novel computationally to further improve the SR performance of single image SR method that learns multiple linear mappings (MLM) based on SR method functions anchored neighborhood regression (ANR) to directly transform LR feature subspaces into HR subspaces. Moreover, we utilize segmentation flipped and rotated with a matrix-vector of the self-examples to expand the internal patch space. Furthermore, we set up MLM from the input LR features to the desire HR outputs to obtain stable. In particular, we improved SR efficiency using optimization principal component analysis (OPCA) based on dimensionality reduction to generate from low-dimensional into high-dimensional with matrix-vector patch space. For our proposed model, there are three regularization parameters analysis that require to optimize the training and refine the outcome. Experimental result indicates that our approach comparison on the standard benchmark with state-of-the-art method validation the effectiveness of our proposed.

Automated segmentation of medical ultrasound (US) images is a challenging problem due to the complicated features of lesions, inconsistent lesions across individuals, and the high segmentation accuracy requirement. From recently published papers in this area, the active contour model (ACM) and machine learning method produce more accurate lesion segmentation results than previous methods. This paper proposes a novel image segmentation approach that integrates an ACM with a generalized linear model (GLM) and forms learning-structured inference. Compared with the GLM, the proposed method can solve the problems of initialization and the local minimum of the ACM. Furthermore, rather than using the ACM as a postprocessing tool, we integrate it into the training phase to fine-tune the GLM. This step allows the use of unlabeled data during training in a semisupervised setting. The integrated model requires only one image as the training set and is not as sensitive to labeled data as other methods. The proposed method is verified using US images, and the results show that the proposed method can produce accurate segmentation results.

In this paper, we present a whole new perspective on generating music. The method proposed in this paper is the first to be used which uses the frequency domain characteristics of pitch contour curve to generate music melody with long-term structure controllable. The music generated by this method has a good long-term structure that other basic music generation methods do not have. This method has great development potential and application ability, can be combined with other music generation methods, and improve the performance of long-term structure. This method firstly uses the neural network to fit the pitch contour curve in frequency domain, then combines the vocabulary matching method to perfect the detailed characteristics of melody generated in time domain and control the long-term trend of notes generated with respect to label information, finally generates music melody with real and controllable long-term structure. Through a large number of experiments, it can be seen that compared with the music generated based on the LSTM, the music generated by the proposed method has better long-term structure and has similar statistical characteristics.

In the face of remote sensing images with diversified information, the recognition of remote sensing images only through local features or global features is limited. Traditionally, it is difficult to achieve good image modeling. To solve this problem, this paper proposes a recognition framework based on dual-channel deep learning(DCDL). The purpose of this framework is to mine global feature information and local feature information at the same time. On the first channel, this paper uses a multi-scale convolution residual network to perform local mining and residual calculations on the image to generate local features; secondly, the local attention mechanism is used to assign weight coefficients to the local features to make the information more contained. More features are more prominent; finally, after several times of mining features and assigning weights, more representative deep features are produced. On the other channel, we introduce the global attention mechanism to realize the weight coefficient distribution of global features; then use the multi-scale dilated convolution to expand the receptive field to obtain a larger range of feature information; then, use the Sigmoid function to achieve 0 to 1 for all features The weight distribution of, further expands the difference between global features; finally, the deep mining of features is realized through 2-layer convolution. In this paper, through the experimental results of the three sub-data sets in the NWPU-RESISC45 data set, we can see that our proposed algorithm has achieved higher recognition accuracy.

Extreme learning machine (ELM) is a training algorithm for single hidden layer feedforward neural network (SLFN), which converges much faster than traditional methods and yields promising performance. In this paper, we hope to present a comprehensive review on ELM. Firstly, we will focus on the theoretical analysis including universal approximation theory and generalization. Then, the various improvements are listed, which help ELM works better in terms of stability, efficiency, and accuracy. Because of its outstanding performance, ELM has been successfully applied in many real-time learning tasks for classification, clustering, and regression. Besides, we report the applications of ELM in medical imaging: MRI, CT, and mammogram. The controversies of ELM were also discussed in this paper. We aim to report these advances and find some future perspectives.

Omnidirectional video quality assessment (OVQA) helps to evaluate the viewers’ visual experience and promotes the development of omnidirectional video. The perceived quality of omnidirectional video is affected not only by the video content and distortion, but also by the viewing directions of viewer’s preference. At present, there are some quality assessment methods for omnidirectional video, but most of them are full-reference (FR). Compared with the FR method, the no-reference (NR) method becomes more difficult due to the lack of the reference video. In this paper, a NR OVQA based on generative adversarial networks (GAN) is proposed, which is composed of the reference video generator and the quality score predictor. Generally, a reference image/video is distorted by some distortion types, and each distortion type has some distortion levels. To the best of our knowledge, there are some NR methods using GAN to generate the reference images/videos for quality assessment. In these methods, the reference images/videos is generated by GAN when the distorted images/videos, which are from a distortion type but with different distortion level, are input into the GAN. In order to achieve an accurate quality assessment, the generated reference images/videos, which are from a distortion type but with different distortion level, are expected to have as similar quality as possible with each other. However, the distorted images/videos are independent for GAN, and the GAN will generate a little different reference images/videos for these distorted images/videos. This issue is not considered in the existing GAN-based methods. To solve this issue, we introduced a level loss in OVQA. For the quality score predictor, as a further contribution of this paper, the viewing direction of the omnidirectional video is incorporated to guide the quality and weight regression. The publicly available dataset is used to evaluate the proposed method. The experimental results indicate the effectiveness of the proposed method.

The innovation of convolutional neural networks motivated the study on the many aspects of image processing and object detection. One topic of interest in the field of object detection is human detection and the human pose estimation scheme. Human pose estimation scheme enables computer vision with a higher accuracy of identifying human motion and movement. Pose estimation scheme has been applied in making 2D images into 3D representation. This paper aims to determine the distance between the camera and the human subject with real-time application which takes advantage of existing pose estimation schemes. The distance scaling applied in the paper showed high performance of about 0.27 m error from the actual distance.

Stock market is a dynamic and volatile market that is considered as time series data. The growth of financial data exposed the computational efficiency of the conventional systems. This paper proposed a hybrid deep learning model based on Long Short- Term Memory (LSTM) and Artificial Bee Colony (ABC) algorithm. ABC is best fit for hyper parameter selection for deep LSTM models and maintains the equilibrium of exploitation and exploration issues. Handling a large volume of multidimensional reviews from social media is a major challenge. This paper evolves the multiple aspects of market sentiments and uses the reliable Big data platform Hadoop ecosystem and its services to compute sentiment polarity index. The ABC-LSTM hybrid model is validated with other core and hybrid models with evolutionary algorithms as Differential Evolution (DE) and Genetic Algorithm (GA). For the experiments, 10 years of historical datasets and social media reviews of IT sector funds Apple Inc. (AAPL), Microsoft corporation (MSFT) and Intel corporation (INTL) from NASDAQ GS, an American stock exchange are considered to validate hybrid forecasting models. Proposed algorithm ABC-LSTM is used to tune the hyperparameters (window size, LSTM units, dropout probability, epochs, batch size and learning rate) and evaluated through Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) as loss function. Performance analysis proves that with sentiment polarity, ABC optimized LSTM obtains improved forecasting accuracy over its counterpart models.

CNN(Convolutional Neural Networks) is a hot topic in the field of pattern recognition., especially in the field of image recognition. And ResNet(Residual Networks) is a special kind of CNN. Compared with the general CNN structure, ResNet introduces the residual unit with an identity mapping. Identity mapping allows the deep layers to directly learn the data received by the shallow layers, which reduces the difficulty of network convergence to a certain extent. As a result, ResNet has a better learning ability, has achieved good performance in various types of image recognition work. The essence of the residual network is to fuse two types of features from different receptive fields, using the fused features instead of the output features of the previous layer as the learning object. But the implementation of feature fusion in original ResNet is adding the two features with equal weights. And this method ignores the fact that the contribution of features from different levels to the learning of the network may not be the same. In this paper, we introduce a feature-inspired gating strategy in the residual unit of ResNet, which allows the network giving different weights to different features, so that the implementation of the feature fusion can be transformed from adding features with equal weights into weighted summation with different weights. And through experiments, we proved that ResNet with gating strategy proposed in this paper can obtain higher recognition accuracy than original ResNet.

In this paper, SFF (Segmentation Facilitate Feature) technique is proposed to find the junction path to segment touched components based on the seed pixel selected among candidate pixels. Handwritten Recognition system has number of applications like reading postal address, filling forms, reading bank cheques, offering several challenges. In practice, constitute of the word images get touched in handwritten data due to variability in stroke, shortage of space which make the individual character extraction from the word image more complicated. Segmentation of individual in a word image requires a technique that takes care of the variability of writing. This paper proposed the SFF (Segmentation Facilitate Feature) technique to find seed pixel among candidate pixels based on 3-neighbouring pixels. It is used to find junction pixels which form a junction path to segregate the touched component. The junction path is selected to avoid the issues arising due to artifacts or deletion of components features. For experimentation, 1840 legal amount words containing touching components are used. The above number includes 250 words from benchmark database (ICDAR) and 1590 words are gathered from 15 different writers. On implementing, SFF (Segmentation Facilitate Feature) technique on the above mentioned database, 89.9% accuracy is achieved and a higher accuracy level 96.2% is achieved when performed on 1000 words containing two touching consonants.

Due to the automatic feature extraction procedure via multi-layer nonlinear transformations, the deep learning-based visual trackers have recently achieved a great success in challenging scenarios for visual tracking purposes. Although many of those trackers utilize the feature maps from pre-trained convolutional neural networks (CNNs), the effects of selecting different models and exploiting various combinations of their feature maps are still not compared completely. To the best of our knowledge, all those methods use a fixed number of convolutional feature maps without considering the scene attributes (e.g., occlusion, deformation, and fast motion) that might occur during tracking. As a pre-requisition, this paper proposes adaptive discriminative correlation filters (DCF) based on the methods that can exploit CNN models with different topologies. First, the paper provides a comprehensive analysis of four commonly used CNN models to determine the best feature maps of each model. Second, with the aid of analysis results as attribute dictionaries, an adaptive exploitation of deep features is proposed to improve the accuracy and robustness of visual trackers regarding video characteristics. Third, the generalization of proposed method is validated on various tracking datasets as well as CNN models with similar architectures. Finally, extensive experimental results demonstrate the effectiveness of proposed adaptive method compared with the state-of-the-art visual tracking methods.

In the COVID-19 pandemic, telehealth plays a significant role in the e-healthcare. E-health security risks have also risen significantly with the rise in the use of telehealth. This paper addresses one of e-health’s key concerns, namely security. Secret sharing is a cryptographic method to ensure reliable and secure access to information. To eliminate the constraint that in the existing secret sharing schemes, this paper presents Tree Parity Machine (TPM) guided patients’ privileged based secure sharing. This is a new secret sharing technique that generates the shares using a simple mask based operation. This work considers addressing the challenges presents in the original secret sharing scheme. This proposed technique enhances the security of the existing scheme. This research introduces a concept of privileged share in which among k number of shares one share should come from a specific recipient (patient) to whom a special privilege is given to recreate the original information. In the absence of this privileged share, the original information cannot be reconstructed. This technique also offers TPM based exchange of secret shares to prevent Man-In-The-Middle-Attack (MITM). Here, two neural networks receive common inputs and exchange their outputs. In some steps, it leads to full synchronization by setting the discrete weights according to the specific rule of learning. This synchronized weight is used as a common secret session key for transmitting the secret shares. The proposed method has been found to produce attractive results that show that the scheme achieves a great degree of protection, reliability, and efficiency and also comparable to the existing secret sharing scheme.

Image encryption mechanisms provide confidentiality and concealment of information (image) in transmission over the alleyway, susceptible to prevalent invasions. With escalating threats in cybersecurity, various cryptography techniques were projected by researchers. The abundance of such mechanisms requires systematic investigation so that that appropriate method can be selected for diverse applications. An efficient encryption technique must analyze all the parameters in an ideal situation and practical cases. Numerous survey papers available in the literature for experimental comparison are devoid of many probable attacks such as anti occlusion attack, chosen-plaintext attack, jpeg compression, etc. This paper provides a qualified study of almost all basic, traditional, chaotic, lightweight, quantum, fractal, and qubit based encryption methods under the influence of prevalent intimidation (Salt & Pepper, Gaussian, Poisson, Rotation attack, chosen-plaintext attack, known-plaintext attack, and so on). We thus carried out an experimental and theoretical investigation based on statistical, differential, and quantitative analysis. To measure the efficacy, all the mechanisms are implemented in MATLAB-2014 and provide the standard deviation to each metric. It is observed that the Quantum and Qubit based algorithms are superlative in comparison to others under the majority of extensive threats due to their sensitive behaviour towards initial conditions and highly random behaviour.

Emotion recognition based on Electroencephalogram (EEG) has attracted much attention in brain-computer interaction. However, most existing methods usually focus on amplitude and spectrum of the EEG signal, leading to sub-optimal performances due to the insufficiency in modelling the complex intrinsic information of neural integration. To address this issue, this paper proposes to capitalize on the largely neglected phase synchronization (PS) between EEG channels which reflects the intrinsic rhythmic interactions between different channels in neural integration. Specifically, this paper develops a simple and novel feature extraction method which calculates the PS based sparse representation features to analyze emotion states. First, the EEG phase synchronization indexes (PSI) of all channel pairs are estimated as features to distinguish different emotions, since certain topographical maps on PSI reveal specific emotion states. Then principal component analysis is performed to eliminate redundant and noisy features in PSI. Finally, Sparse Representation based Classification (SRC) furtherly emphasize emotion-related features and restrain useless features. For the benchmark affective EEG dataset DEAP, the proposed method based on no-overlapping EEG features achieve an average accuracy of 94.5%, 87.61%, and 67.04% for the classification tasks respectively on two, three and four emotions, demonstrating the superiority over state-of-the-art emotion classification methods.

Devanagari script is the most widely used script in India and other Asian countries. There is a rich collection of ancient Devanagari manuscripts, which is a wealth of knowledge. To make these manuscripts available to people, efforts are being done to digitize these documents. Optical Character Recognition (OCR) plays an important role in recognizing these documents. Convolutional Neural Network (CNN) is a powerful model that is giving very promising results in the field of character recognition, pattern recognition etc. CNN has never been used for the recognition of the Devanagari ancient manuscripts. Our aim in the proposed work is to use the power of CNN for extracting the wealth of knowledge from Devanagari handwritten ancient manuscripts. In addition, we aim is to experiment with various design options like number of layes, stride size, number of filters, kenel size and different functions in various layers and to select the best of these. In this paper, the authors have proposed to use deep learning model as a feature extractor as well as a classifier for the recognition of 33 classes of basic characters of Devanagari ancient manuscripts. A dataset containing 5484 characters has been used for the experimental work. Various experiments show that the accuracy achieved using CNN as a feature extractor is better than other state-of-the-art techniques. The recognition accuracy of 93.73% has been achieved by using the model proposed in this paper for Devanagari ancient character recognition.

Currently, embedded systems can be found everywhere in quotidian life. In the development of embedded systems, information security is one of the important factors. Encryption is an efficient technique to protect information against attacks. However, because of constraints, existing encryption functions are not compatible and do not agree with real-time applications in embedded systems. In this paper, an improved cryptographic approach with a high level of security and high speed is put forward. Our work uses an efficient version of a hybrid scheme comprising an Advanced Encryption Standard (AES) - Elliptic Curve Cryptography (ECC) for medical image encryption, which combines the benefits of the symmetric AES to speed-up data encryption and asymmetric ECC in order to secure the interchange of a symmetric session key. The contribution of this paper consists of the following two main points: First, we put forward an optimized ECC hardware architecture to respect the compromise between area, power dissipation, and speed. Thus, we primarily utilize only two multipliers to develop the Point Addition (PA) block and the Point Doubling (PD) block, which reduces time complexity. Then, a 32-bit multiplier and a 32-bit inverter architecture based on shifts and XORs are proposed to reduce power consumption and area occupancy. Second, for image encryption, we primarily propose to modify the AES by eliminating the mix-columns transformation and replacing it with a permutation based on the shifts of columns, which decreases time complexity while maintaining the Shannon diffusion and the confusion principle. Then, an adjustment of the rearrangement of the general structure is given to enhance the entropy value. The global cryptosystem is implemented using a co-design approach where the modified AES runs on the NIOS II processor, and the scalar ECC multiplication is designed as a hardware accelerator. The suggested cryptographic system spends much less execution time, which is a significant factor for being applied in practice. Security analysis is successfully performed, and our experiments prove that our proposed technique provides the basics of cryptography with more simplicity and correctness. In fact, the results of the evaluation prove the effectiveness, rapidity and high security of the suggested algorithm.

The prediction of electricity consumption is a vital foundation for smart energy management. Since the consumption of power varies with different appliances, better forecasting of power and peak demand is an essential accomplishment for the proper planning and development of the power generation and distribution system. This forecast analysis helps the service providers and the government to understand the lifestyle of the customers. The existing prediction and forecasting models are not meeting the standard requirements and moreover difficult to apply in practice. The forecast says that the boom of electric vehicles will increase the demand of electricity globally by 3% for the upcoming year. There exists a number of machine learning algorithms for classification and decision making. But the accuracy of the exiting methods have shown inferior performance in terms of prediction which leads to inefficient decision making in the quantity of electricity generation. This paper proposes the use of random forest supervised learning model to forecast the consumption of power and identify the level of peak demand. The large smart meter dataset collected at varying seasons of the year is fed to the random forest classifier technique for better analysis and forecasting. This approach outperforms in terms of accuracy, stability and generalization. In addition, this paper investigates the existing models and compares the performance with those models. The performance analysis shows that this model performs better than the other investigated models with performance accuracy of 95.67% and enhanced accuracy of precision and recall.

Activity recognition represents the task of classifying data derived from different sensor types into one of predefined activity classes. The most popular and beneficial sensors in the area of action recognition are inertial sensors such as accelerometer and gyroscope. Convolutional neural network (CNN) as one of the best deep learning methods has recently attracted much attention to the problem of activity recognition, where 1D kernels capture local dependency over time in a series of observations measured at inertial sensors (3-axis accelerometers and gyroscopes) while in 2D kernels apart from time dependency, dependency between signals from different axes of same sensor and also over different sensors will be considered. Most convolutional neural networks used for recognition task are built using convolution and pooling layers followed by a few number of fully connected layers but large and deep neural networks have high computational costs. In this paper, we propose a new architecture that consists solely of convolutional layers and find that with removing the pooling layers and instead adding strides to convolution layers, the computational time will decrease notably while the model performance will not change or in some cases will even improve. Also both 1D and 2D convolutional neural networks with and without pooling layer will be investigated and their performance will be compared with each other and also with some other hand-crafted feature based methods. The third point that will be discussed in this paper is the impact of applying fast fourier transform (FFT) to inputs before training learning algorithm. It will be shown that this preprocessing will enhance the model performance. Experiments on benchmark datasets demonstrate the high performance of proposed 2D CNN model with no pooling layers.

This paper presents a novel encryption scheme based on genetic operations and Mixed Linear-Nonlinear Coupled Logistic Map Lattice (MLNCML) spatiotemporal chaotic system. The initial values of the MLNCML system are related to the plain-image. Therefore, this scheme is sensitive to the plaintext and keys. For the simplicity and efficiency of the encryption process, expanded XOR (eXOR) operation and genetic operations are applied in the scheme. Spatiotemporal chaotic system can resist dynamical degradation and periodic phenomena, which offers good randomness for encryption. Moreover, the eXOR operation in this paper can effectively avoid the chosen-plaintext attack of traditional XOR operation. The DNA mutation and crossover operations are controlled by chaotic sequences, which overcome man-in-the-middle attack in the traditional DNA addition and subtraction operation. The experimental results indicate that the proposed scheme has good security to resist common attacks.

Due to the huge flow of data and complications in mutable characteristics of the data, Distributed Denial of Services attacks existed in the Multimedia Internet of Things. Attacks over the IoT have become an increasing menace in recent time, which tries to hack or illegally tamper the streaming data available over the networks. On the other hand, there has been an increase in volume in research contributions to effectively counter these attacks and implement a strong defense mechanism. There have been numerous algorithms and frameworks implemented in recent times that are intelligent and soft computing-based. These evolution-based algorithms play a vital role in self-adapting the system under attack towards increasing and new types of attacks which are increasing day by day. One such area of soft computing algorithms investigated in this paper is the Artificial Neural Network or popularly known as ANNs. It works analogously to the biological neurons in the human body. In this paper, we systematically explain the ANN-based network model to counteract the DDoS attacks in the Multimedia Internet of Things, architecture, and implementation of ANNs, the experimental investigations and findings which help in drawing an inference of ANN-based defense models.

Matching images of a scene with high precision is a core component in real time applications such as image-based navigation of unmanned aerial vehicles, terminal guidance of missiles. The aim of the research discussed in this paper is to develop a scene matching algorithm of high accuracy, robust to the variations between the images compared with scope for parallelism. After a study of state-of-the-art scene matching algorithms, a local feature-based approach was chosen to meet the stated requirements. The paper discusses the techniques for modules of the algorithm chosen and the methodology implemented after a metric study with a dynamic threshold based nearest neighbour distance ratio matching strategy (DT-NNDR). The metrics and the proposed strategy were tested for the matching of planar images from Affine covariant features group, unmanned aerial vehicle videos from a national defence agency, and satellite images from Google. The performance of strategy was compared with the existing NN-DR strategy with an experimental optimal threshold. The overall precision of the proposed strategy is 88%, making it suitable for applications which require high value of true positive.

Object recognition is a key research area in the field of image processing and computer vision, which recognizes the object in an image and provides a proper label. In the paper, three popular feature descriptor algorithms that are Scale Invariant Feature Transform (SIFT), Speeded Up Robust Feature (SURF) and Oriented Fast and Rotated BRIEF (ORB) are used for experimental work of an object recognition system. A comparison among these three descriptors is exhibited in the paper by determining them individually and with different combinations of these three methodologies. The amount of the features extracted using these feature extraction methods are further reduced using a feature selection (k-means clustering) and a dimensionality reduction method (Locality Preserving Projection). Various classifiers i.e. K-Nearest Neighbor, Naive Bayes, Decision Tree, and Random Forest are used to classify objects based on their similarity. The focus of this article is to present a study of the performance comparison among these three feature extraction methods, particularly when their combination derives in recognizing the object more efficiently. In this paper, the authors have presented a comparative analysis view among various feature descriptors algorithms and classification models for 2D object recognition. The Caltech-101 public dataset is considered in this article for experimental work. The experiment reveals that a hybridization of SIFT, SURF and ORB method with Random Forest classification model accomplishes the best results as compared to other state-of-the-art work. The comparative analysis has been presented in terms of recognition accuracy, True Positive Rate (TPR), False Positive Rate (FPR), and Area Under Curve (AUC) parameters.

The influence of Internet of Things (IoT) is growing exponentially in the consumer patterns and will continue to increase in the coming years. With millions of interconnected devices over the internet, IoT is currently running into its monumental security and centralization issues that can be resolved using Blockchain. This paper aims to demonstrate the use of a smart contract on a private Ethereum Blockchain (also known as permission less Blockchain) to check whether a user holds enough tokens to utilize a service. Private in this context implies a private instance of the Ethereum implementation in which we make use of the same security and principles of the Ethereum main chain on a personal network of connected IoT devices which improves the overall privacy and security of the system than the traditional schemes. The proposed model consists of a target IoT device controlled by a Raspberry Pi, running a client application that makes decisions based on the state of the smart contract. With this environment, we are able to achieve the following goals: (1) no data leaves the private network (2) all services are employed with transparency and accountability (3) only registered devices can issue requests for using a service. Although the demonstration is made on a home network, the model presented in this paper can be utilized in commercial environments where any service can be availed from a wide array of smart devices through a mobile application.

In this paper, we recognize serial numbers on banknotes using deep learning. The samples used in this paper are digital images which have been preprocessed with data labelling and data augmentation, e.g., scaling transformation, etc. The algorithms based on deep learning are proposed and have the stability for serial number recognition with complex backgrounds. In this paper, a pipeline of deep neural networks is established for the recognition of banknote serial numbers. Because high reliability is more important than accuracy in financial applications, DenseNet is set forth as the primary classifier, the scaling transformation of SegLink is put forward to locate the characters, the detection rate is up to 95.80%. A convolutional neural network with residual attention model is proposed for serial number recognition, the precision is up to 97.09%.

In this paper, a chaos-based neural synchronization has been proposed for the development of the public-key exchange protocol. A special neural network structure called Tree Parity Machine (TPM) is used for neural synchronization. Two TPMs accept the common input and different weight vector and update the weights using neural learning rule by exchanging their output. In some steps, it results in complete synchronization, and the weights of the two TPMs become identical. These identical weights serve as a secret key. There is, however, hardly some investigation to investigate the randomness of the common input vector used in the synchronization process. In this paper, logistic Chaos system based Tree Parity Machine (CTPM) is proposed. For faster synchronization, this proposed CTPM model uses logistic chaos generated random common input vector. This proposed CTPM model is faster and has better security than TPM with the same input, output, and hidden neurons. This proposed technique has been passed through a series of parametric tests. The results have been compared with some recent techniques. The results of the proposed technique have shown effective and robust potential.

User-Generated Content (UGC) is turning into the predominant type of internet traffic. Content popularity prediction plays a pivotal role in managing this large-scale traffic. As a result, popularity prediction is increasingly becoming an important area of research in computer networking. Generally, popularity prediction methods are classified into two groups, namely, feature-driven and early-stage. While feature-driven methods predict content popularity before publication, early-stage methods monitor early content popularities to forecast the future. Many papers have shown that early-stage popularity prediction performs better than feature-driven methods. In this paper, we improve the performance of early-stage popularity prediction by first, classifying the data into several clusters using k-means clustering with Pearson correlation distance, and then, training a Deep-Belief Network (DBN) for each cluster. We evaluate our method using a dataset of YouTube videos and show that using a generative model such as DBN for time series prediction significantly improves the performance. Numerical results indicate that our proposed method outperforms other state-of-the-art methods by reducing Mean Absolute Percentage Error (MAPE) and mean Relative Square Error (mRSE) by up to 47.86% and 25.18%.

Soil classification is one of the major affairs and emanating topics in a large number of countries. The population of the world is rising at a majorly rapid pace and along with the increase in population, the demand for food surges actively. Typical techniques employed by the farmers are not adequate enough to fulfill the increasing requirements and therefore they have to hinder the cultivating soil. For proper crop yield, farmers should be aware of the correct soil type for a particular crop, which affects the increased demand for food. There are various laboratory and field methods to classify soil, but these have limitations like time and labor-consuming. There is a requirement of computer-based soil classification techniques which will help farmers in the field and won’t take a lot of time. This paper talks about different computer-based soil classification practices divided into two streams. First is image processing and computer vision-based soil classification approaches which include the conventional image processing algorithms and methods to classify soil using different features like texture, color, and particle size. Second is deep learning and machine learning-based soil classification approaches, such as CNN, which yields state-of-the-art results. Deep learning applications mostly diminish the dependency on spatial-form designs and preprocessing techniques by facilitating the end-to-end process. This paper also presents some databases created by the researchers according to the objective of the study. Databases are created under different environmental and illumination conditions, using different appliances such as digital cameras, digital camcorder, and a smartphone camera. Also, evaluation metrics are briefly discussed to layout some graded measures for differentiation. This review serves as a brief guide to new researchers in the field of soil classification, it provides fundamental understanding and general knowledge of the modern state-of-the-art researches, in addition to skillful researchers considering some dynamic trends for future work.

Camouflaged communication using a media is known as Steganography. It is different than encryption as the presence of message is also concealed in case of steganography. The message however can be encrypted before hiding in a media. Detection of concealed exchange being carried out or unraveling the details of such transmission is known as Steganalysis. Steganalysis can be detected by classifying the given media file as cover media file or stego media file. Blind steganalysis detects presence of hidden content without any knowledge about the cover media file and steganography algorithm used. Steganalysis plays a vital role in forensics of various media such as text, audio, image, video and network packets. Machine learning techniques have been widely used for steganalysis in literature. These techniques use a three step approach consisting of Feature Extraction, Training and Testing Phases. Deep learning techniques, a subset of machine learning techniques are preferred by researchers over machine learning techniques as (i) they consist of Training and Testing Phases with the feature extraction step done automatically, (ii) they give better accuracy when trained with huge amount of data. This paper proposes novel multi class blind steganalysis technique for images. Convolutional Neural Network (CNN) is one of the best known architecture used with image steganalysis. But as the depth of CNN architecture increases, problem of vanishing descent arise which affects the accuracy. In order to solve the problem of the vanishing/exploding gradient in CNN, concept called Residual Network which use a technique called skip connections is being used. The skip connection skips training from few layers and connects directly to the output. A deep residual network helps to automatically capture complex statistical features of images and preserve weak stego signal in image content making it suitable for multi class blind steganalysis. This paper uses deep residual network for multi-class blind steganalysis. Proposed DRN has been successfully applied for multi class blind steganalysis in spatial and JPEG images. Experimental results demonstrate proposed network is comparable to state or art techniques present in literature.

In this paper, a new scrambling algorithm is proposed to reduce the correlation between the image pixels in three directions. In this paper, this algorithm is called ladder scrambling, and the image is scrambled multiple times in the form of diagonal lines. In terms of diffusion, DNA coding is used, and the secondary processing of the image is performed through DNA coding to change the pixel value of the image. The final encrypted image is affected by three conditions: the number of scrambling algorithms, the selection of DNA coding rules, and the key of the chaotic system. Theoretical analysis and experimental results show that the algorithm is safe and effective.

Recently, data hiding techniques have become very popular in several vital applications, especially in telemedicine. The reason for this is their ability to give good results such as high embedding capacity while preserving visual image quality as much as possible after extracting the hidden secret message. In earlier studies, many researchers have achieved the goal of reversible data hiding (RDH) algorithm. All these methods have achieved excellent results on standard and natural images. However, in the case of medical images, especially color medical images, we face the problem of how to preserve the visual quality of image contents while achieving the goals of RDH in avoiding the loss of patient data or the distortion of the diagnosing image. In this paper, we proposed a secure data hiding method using a hyper chaotic map and left-most embedding strategy. The proposed methods are hybrid, where it is applied in the DCT frequency domain and encrypted domain together as presented here. This gives a higher embedding rate and higher visual image quality than existing methods without any loss or distortion of both hidden secrets data and reconstructed image. The novelty of this paper is to embed the desired secret data in each quantized block of DCT using (8-bit LMSB) strategy for embedding process. We tested our algorithm on both color medical images and standard color images of different sizes and different formats. We evaluated the performance of our algorithm on the basis of the quality metrics MSE, PSNR, BER, SSIM, Correlation, Symbol Error Rate, additional quality evaluation metrics, execution time, and different types of geometric and signal attacks. All of these parameters are demonstrated and represented in this proposed work in detail.

The main focus of this paper is the separation of underdetermined convolutive blind speech in a multi-speaker environment. We present a method based on mask prediction in the time-frequency domain. Firstly, depending on the sparsity of signals in the time-frequency (TF) domain, we extimate speakers’ masks by clustering the relative absolute and Hermitian angle features extracted from the frequency components of the mixtures. Speech separation algorithms that are based on the sparsity and disjoint orthogonality of the speech signals in the time-frequency domain are not efficient when more than one source is active. Hence, in this paper, the cluster centers are estimated mostly based on the TF units that probably have only one active source. The correlations between the estimated masks, belonging to adjacent frequency bins, are leveraged to solve the permutation problem. To increase the accuracy, we have zeroed the value of masks at the TF unit without any active source. Moreover, in clustering, we employ a weighting function to consider the parts of masks that probably contains just one active source. Finally, in order to decrease the musical noise of the separated signals and improve their quality, sparse filters in the time-domain are utilized to re-estimate the separated signals. Performance of the proposed method is evaluated by a number of simulated and real speech signals. The simulated experiments were performed using a public dataset and Roomsim simulator. Compared the proposed method with some conventional algorithms, we observed that our separation method is more accurate than other approaches.

Video compression makes the encoded video stream more vulnerable to the channel errors so that, the quality of the received video is exposed to severe degradation when the compressed video is transmitted over the error-prone environments. Therefore, it is necessary to apply error concealment (EC) techniques in the decoder to improve the quality of the received video. In this regard, an Adaptive Content-based EC Approach (ACBECA) is proposed in this paper, which exploits both the spatial and temporal correlations within the video sequences for the EC purpose. The proposed approach adaptively utilizes two EC techniques, including new spatial-temporal error concealment (STEC) technique, and a temporal error concealment (TEC) technique, to recover the lost regions of the frame. The STEC technique proposed in this paper is established on the basis of non-Local Means concept and tries to recover each lost macroblock (MB) as the weighted average of the similar MBs in the reference frame, whereas the TEC technique recovers the motion vector of the lost MB adaptively by analyzing the behavior of the MB in the frame. The decision on temporally or spatially reconstructing the degraded frames is made dynamically according to the content of the degraded frame (i.e., structure or texture), type of the error and also block loss rate (BLR). Compared with the state-of-the-art EC techniques, the simulation results indicate the superiority of the ACBECA in terms of both the objective and subjective quality assessments.

In this paper, we propose a new method for improving human activity recognition (HAR) datasets in order to increase their classification accuracy when trained with a certain classifier like a Neural Network. In this paper a novel training/testing process for building/testing a classification model for human activity recognition (HAR) is proposed. Traditionally, HAR is done by a classifier that learns what activities a person is doing by training with skeletal data obtained from a motion sensor such as Microsoft Kinect or accelerometer sensors. These skeletal data are the spatial coordinates (x, y, z) of different parts of the human body. In addition to the spatial features that describe current positions in the skeletal data, new features called Shadow Features are used to improve the supervised learning efficiency and accuracy of Neural Network classifiers. Shadow Features are inferred from the dynamics of body movements, thereby modelling the underlying momentum of the performed activities. They provide extra dimensions of information for characterizing activities in the classification process and thus significantly improving the accuracy. These Shadow Features are generated based on the existing features obtained from sensor datasets. In this paper we show that the accuracy of a neural network classifier can be significantly improved by the addition of Shadow Features and we also show that the generation of Shadow Features can be achieved with little time cost, on the fly, with the NVIDIA GPU technology and the CUDA programming model, this way we can improve the Neural Network accuracy at almost no time cost. GPUs are particularly suitable for generating Shadow Features, since they possess multiple cores which can be taken advantage of, in order to generate Shadow Features for multiple data columns in parallel, therefore reducing a lot of processing time, especially when dealing with huge datasets.

This research paper presents a comparative study between 14 state of art descriptors which includes Local Binary Pattern (LBP), Median Binary Pattern (MBP), 6 × 6 Multiscale Block LBP (6 × 6 MB-LBP), Local Neighborhood Difference Pattern (LNDP), Logically Connected-LBP (LC-LBP), Local Phase Quantization (LPQ), Compound LBP (CLBP), Horizontal Elliptical LBP (HELBP), Vertical Elliptical LBP (VELBP), ELBP, Neighborhood Intensity Based LBP (NI-LBP), Median Robust Extended LBP Based on NI (MRELBP-NI), Radial Difference-LBP (RD-LBP) and Transition LBP (tLBP). For all the descriptors the features are extracted globally and the dimensionality of the feature size is reduced by employing Principal Component Analysis (PCA) and Fishers Linear Discriminant Analysis (FLDA). Finally classification is performed by Support Vector Machines (SVMs) and Nearest Neighbor (NN). Experiments are performed on 8 challenging databases which covers all the major challenges such as pose variations, illumination variations, expression variations and occlusion changes. The 8 challenging databases includes ORL, GT, Faces94, MIT-CBCL, Yale, YB, EYB and SOF. Out of all the descriptors it is the performance of the CLBP descriptor which is most encouraging. On some occasions the MRELBP-NI descriptor also achieves good results. But all in all the CLBP descriptor achieves the best results. In addition to this Deep learning based descriptors are also discussed in the paper.

Over the last few years, there has been a rapid growth in digital data. Images with quotes are spreading virally through online social media platforms. Misquotes found online often spread like a forest fire through social media, which highlights the lack of responsibility of the web users when circulating poorly cited quotes. Thus, it is important to authenticate the content contained in the images being circulated online. So, there is a need to retrieve the information within such textual images to verify quotes before its usage in order to differentiate a fake or misquote from an authentic one. Optical Character Recognition (OCR) is used in this paper, for converting textual images into readable text format, but none of the OCR tools are perfect in extracting information from the images accurately. In this paper, a method of post-processing on the retrieved text to improve the accuracy of the detected text from images has been proposed. Google Cloud Vision has been used for recognizing text from images. It has also been observed that using post-processing on the extracted text improved the accuracy of text recognition by 3.5% approximately. A web-based text similarity approach (URLs and domain name) has been used to examine the authenticity of the content of the quoted images. Approximately, 96.26% accuracy has been achieved in classifying quoted images as verified or misquoted. Also, a ground truth dataset of authentic site names has been created. In this research, images with quotes by famous celebrities and global leaders have been used. A comparative analysis has been performed to show the effectiveness of our proposed algorithm.

Content-based Image Retrieval (CBIR) is a technique that can exploit the wealth of the data stored in a repository and help radiologists in decision making by providing references to the image in hand. A CBIR system for High-Resolution Computed Tomography (HRCT) lung images depicting signs of Interstitial Lung Diseases (ILDs) can be built and used as a self-learning tool for budding radiologists. The study of a few lung image retrieval systems available in the literature identifies some important issues that need to be taken care of. In most of the works, the creation of the reference database involves painstaking manual activity, which is time-consuming and needs skilled labor. A lot of human interventions are required, particularly for the proper delineation of the region of interest (ROI) that represents pathology in each of the images in a database. In most cases, the size of the ROIs representing different disease findings are fixed (i.e., either a fixed size square or circle), which at times may not be a proper representation of the disease pattern and as a consequence, it might limit the system’s performance. Until date, a few learning-based approaches have been developed for content-based image retrieval of HRCT lung images, which either learn the similarity using a classifier or get trained through relevance feedback. For medical image analysis, the availability of labelled data for learning makes these learning-based retrieval systems meaningful as it enhances their performance in contrast to their simple distance-based counterpart. The objective of this paper is to develop a CBIR system for ILDs that is reliable and needs minimal human intervention. The paper evaluates the performance of three popular segmentation algorithms. It identifies the best for the effective and automated delineation of an arbitrary region of interest (AROI) depicting the sign of ILDs on HRCT images of the thorax in contrast to the manual delineation of fixed size ROI. This minimizes the manual effort for the creation and maintenance of the reference database, as well as the manual delineation of AROI during query formation. Moreover, AROI created through the automated clustering is found to have a better representation of disease patterns. Three recently proposed general-purpose learning based CBIR techniques are implemented and tested for retrieval of HRCT lung images depicting the sign of ILDs. The best method is suggested after careful evaluation of all the competing techniques.

Biometric technologies have become the main focus in the design of state-of-the-art border security solutions. While respective research in the field of multimedia vision has been centred around improving quality and accuracy of identity recognition, the impact of such technologies upon society and legal regulations still remains a topic unaddressed, specifically within the engineering community. Research in technology can and in some respect must include collaboration with social sciences and social practice. Building on participation in the EU funded research project PERSONA [18] (Privacy, Ethical, Regulatory and SOcial No-gate crossing point solutions Acceptance), authors of this paper look at the challenges associated with biometrics-based solutions in no-gate border crossing point scenarios. This included the procedures needed for the assessment of their social, ethical, privacy and regulatory acceptance, particularly in view of the impact on both, the passengers and border control authorities as well as the potential pitfalls of biometric technology due to fraudulent activities. In consultation with the collaborating border control authorities, the paper reports on the formal assessment of biometric technologies for real-world acceptance to cope with the increasing demand of global travellers crossing state borders.

Multimedia security has been extensively used in content protection, image authentication, data hiding and signal encryption. Similarly, transmission of biomedical data or information remotely for healthcare applications should be secure. One of the important medical signals that need to be transmitted to healthcare centers is the Electrocardiogram (ECG) signal. This paper is concerned mainly with ECG signal encryption for security applications. The paper presents three cryptosystems for ECG signal encryption based on the fusion of ECG signals with other masking signals that are rich in activities such as speech signals. The common thread between these cryptosystems is the operation on sample values of the ECG signal rather than adopting encoding and decoding schemes, and this saves much time and is more immune to both noise and hacking scenarios. The proposed cryptosystems are compared to the encryption technique that uses 1-D logistic map. The performances of the proposed cryptosystems are evaluated through simulation experiments in terms of histogram, structural similarity index, Signal-to-Noise Ratio (SNR), log-likelihood ratio, spectral distortion and correlation coefficient. It is clear from the experiments that the utilization of more levels of encryption increases the security.

Text classification could be defined as the way of allocating text into predefined groups according to its contents. Over the past few years, an increase emerged in the volume of information in the varied fields on the Internet, thus making the classification of texts one of the most important, yet challenging. Text classification is commonly employed in numerous applications and for different objectives. The extensive and broad use of the Internet, particularly in the Arab world, as well as the massive number of the documents and pages which are provided in the Arabic language, raised the need for having suitable tools for classification of these pages and documents by their main categories. The aim of this paper to study the effect of the improved CHI (ImpCHI) Square on the performance of six well-known classifiers: Random Forest, Decision Tree, Naive Bayes, Naive Bayes Multinomial, Bayes Net, and Artificial Neural Networks. These proposed techniques are quite important for improving classification of Arabic documents and can be regarded as a promising basis for the stage of text classification because it contributes to the classification of the texts into predefined categories. This combination method takes the advantages of more than one technique, which can produce better results in the final outcomes. The dataset employed in this paper includes 9055 Arabic documents that were collected from various Arabic resources. Based on their content, these documents were divided into twelve categories. Four performance evaluation criteria were used: the F-measure, recall, precision, and Time build model. The experimental results show that the use of ImpCHI square gives better classification results than the normal CHI square method with all studied classifiers, in terms of all used performance criteria.

A deterioration level estimation method via neural network maximizing category-based ordinally supervised multi-view canonical correlation is presented in this paper. This paper focuses on real world data such as industrial applications and has two contributions. First, a novel neural network handling multi-modal features transforms original features into features effectively representing deterioration levels in transmission towers, which are one of the infrastructures, with consideration of only correlation maximization. It can be realized by setting projection matrices maximizing correlations between multiple features into weights of hidden layers. That is, since the proposed network has only a few hidden layers, it can be trained from a small amount of training data. Second, since there exist diverse characteristics and an ordinal scale in deterioration levels, the proposed method newly derives category-based ordinally supervised multi-view canonical correlation analysis (Co-sMVCCA). Co-sMVCCA enables estimation of effective projection considering both within-class divergence and the ordinal scale between classes. Experimental results showed that the proposed method realizes accurate deterioration level estimation.

Based on the n-ary counting system, combined with the matrix semi-tensor product theory and Hilbert curve, a chaotic image encryption algorithm is designed. Different from the traditional encryption method, the algorithm proposed in this paper is an encryption algorithm with scrambling and diffusion at the same time. First, the pixel value is converted from decimal to n-ary. In the n-ary counting system, the plaintext image is randomly divided into some groups, and the Hilbert curve is used for scrambling to each group. The blocks are converted into scrambled images, so that the scrambling and diffusion can be carried out at the same time. Then, in order to improve the security of the algorithm, another round of diffusion is carried out based on matrix semi-tensor product mechanism. Chaotic sequence is generated by Chen system. This chaotic sequence performs matrix semi-tensor product operation with the first round of encrypted image, and generate second encrypted images. Finally, this encryption method is applied to color image encryption. Compared with some representative algorithms, the experimental results show that the algorithm proposed in this paper is secure and it can resist common attacks.

This paper develops an efficient color image compression method based on the DCT and a new color base. Digital color images are commonly represented in RGB space. Generally, it is noted that a strong correlation exists between the three planes R, G, and B of a color image. The reduction of this correlation certainly offers an advantage in the compression of RGB images. In this context, there is an infinite number of possible spaces to represent the RGB image. The main contributions of this paper are summarized in two main points. First, we design an optimized color space B1B2B3 using the Bat algorithm (BA) to pass from the RGB space to space more appropriate for each image. This space is produced by maximizing the energy of the image in the plane B1 more than in B2 and B3. Second, we produce optimized thresholds appropriate to each plane of the converted image. The Bat algorithm optimizes the cost function to compute thresholds to partially reduce the number of the less significant DCT coefficients that correspond to the lower quantity of energy. The reported results against those of recent methods prove that the proposed method presents high performances in terms of peak signal to noise ratio (PSNR) on the commonly used test color images as well as the test medical images in literature.

This paper aims to early Breast Cancer (BC) detection by Mammography (MG) established on the production of excellent images and competent interpretation. This paper proposes two algorithms for Object Detection and Diagnosis (ODD) the BC. The first proposal is relied on merging the features of A Trous Algorithm (AT) with Homomorphic Processing (HP)Using Contrast Limited Histogram Equalization (AHUC) following by Segmentation and Feature Extraction (FE) for Classification (ASFC). The ASFC depends on enhancement using AHUC in addition to pre-processing followed by segmentation using Optimum Global Thresholding (OT) and finally FE for the detection or classification task. The second scheme is based segmentation after Hybrid structure HE and Fuzzy model (HEF) and finally FE for Classification (HEFC) The HEFC depends on improvement using HEF and pre-processing followed by segmentation using OT and finally FE for classification task. The performance quality metrics for the suggested techniques are entropy, average gradient, contrast, Sobel edge magnitude, homogeneity, sensitivity, specificity, precision and accuracy.Simulation results prove that the success of both techniques in detecting the BC objects. By comparing the first and the second presented algorithms, it is clear that the second suggested technique gives superior for the ODD the BC.

Movie genre classification is a challenging task that has increasingly attracted the attention of researchers. The number of movie consumers interested in taking advantage of automatic movie genre classification is overgrowing, thanks to media streaming service providers’ popularization. In this paper, we addressed the multi-label classification of movie genres in a multimodal way. To this end, we created a dataset composed of trailer video clips, subtitles, synopses, and movie posters from 152,622 movie titles of the Movie Database (TMDb). Such a large dataset was carefully curated, organized, and made available as a contribution of this work. We labeled each movie of the dataset according to a set of eighteen genre labels. In the experimental evaluation performed in this paper, we computed different kinds of descriptors, such as Mel Frequency Cepstral Coefficients (MFCCs), Statistical Spectrum Descriptor (SSD), Local Binary Pattern (LBP) from spectrograms, Long-Short Term Memory (LSTM), and Convolutional Neural Networks (CNN). With these descriptors, we trained different monolithic classifiers using BinaryRelevance and ML-kNN techniques. Besides, we also explored the combination of classifiers/features using a late fusion strategy. The fusion of a LSTM trained on synopses and another LSTM trained on the movie subtitles provided our best results in F-Score (0.674) and AUC-PR (0.725) metrics. These results corroborate the existence of complementarity among classifiers trained on different sources of information in this field of application. As far as we know, this is the most comprehensive study developed in terms of diversity of multimedia sources of information to perform movie genre classification.