Search

Introduction: This paper introduces novel and reliable approach for speech impaired people to assist them to communicate effectively in real time. A deep learning technique named as convolution neural network is used as its classifier. With the help of this algorithm, words are recognized from an input which is a visual speech, disregards with its audible or acoustic property. Methods: This network extracts the features from mouth stances and different images respectively. With the help of a source, non-audible mouth stances are taken as an input and then segregated as subsets to get desired output. The Complete Datum is then arranged to recognize the word as an affricate. Results: Convolution neural network is one of the most effective algorithms that extracts features, performs classification and provides the desired output from the input images for speech recognition system. Conclusion: Recognizing the syllables at real time from visual mouth stances input is the main objective of the proposed method. When tested, datum accuracy and quantity of training sets is giving satisfactory output. A small set of datum is taken as first step of learning. In future, large set of datum can be considered for analyzing the data. Discussion: On the basis of type of Datum, network proposed in this paper is tested to obtain its precision level. A network is maintained to identify the syllables but it fails when syllables are of same set. Requirement of Higher end graphics pro-cessing units is there to bring down the time consumption and increases the efficiency of network.

Recently, the deformable derivative and its properties have been introduced. In this work, we have investigated the concept of deformable Laplace transform (DLT) in more detail. Furthermore, some classical properties of the DLT are also included. The Heaviside expansion formula and convolution theorem for deformable inverse Laplace transform are also discussed. Furthermore, some illustrative numerical examples are also discussed to validate the applicability of the proposed DLT and finally conclude the theory.

Mapping human cognition into automated analysis is the key area of research due to its fascinating applications in almost every area of developing artificially intelligent machines. The best way to understand the functioning of brain is to study electroencephalogram (EEG) patterns, therefore, a lot of research has been directed towards studying EEG signals. Since, EEG recordings are subject dependent and exhibit variations due to external influences or type of recording instruments, it is hard to develop a generalized affect categorization system that can provide robust affect labeling to the EEG patterns. To overcome this, proposed work presents a novel general framework for affect-based cognitive analysis. The proposed system involves following steps: pre-processing, feature selection, Generalized Procrustes Analysis (GPA) step to reduce the inter-class and intra-class variance and finally, the processed pattern is passed on to a trained classifier for classifying the pattern into appropriate affect categories. The presented approach has been tested on single as well as multiple subjects’ EEG data taken from two different datasets, Database for Emotion Analysis using Physiological signals (DEAP) and SJTU Emotion EEG Dataset (SEED) and performance of popular classifiers are assessed. The experimental results suggest that SVM classifier is the best among the selected ones for classifying single as well as mixed subjects’ data.

Riccati differential equations (RDEs) plays important role in the various fields of defence, physics, engineering, medical science, and mathematics. A new approach to find the numerical solution of a class of RDEs with quadratic nonlinearity is presented in this paper. In the process of solving the pre-mentioned class of RDEs, we used an ordered combination of Green’s function, Adomian’s polynomials, and Pade` approximation. This technique is named as green decomposition method with Pade` approximation (GDMP). Since, the most contemporary definition of Adomian polynomials has been used in GDMP. Therefore, a specific class of Adomian polynomials is used to advance GDMP to modified green decomposition method with Pade` approximation (MGDMP). Further, MGDMP is applied to solve some special RDEs, belonging to the considered class of RDEs, absolute error of the obtained solution is compared with Adomian decomposition method (ADM) and Laplace decomposition method with Pade` approximation (LADM-Pade`). As well, the impedance of the method emphasised with the comparative error tables of the exact solution and the associated solutions with respect to ADM, LADM-Pade`, and MGDMP. The observation from this comparative study exhibits that MGDMP provides an improved numerical solution in the given interval. In spite of this, generally, some of the particular RDEs (with variable coefficients) cannot be easily solved by some of the existing methods, such as LADM-Pade` or Homotopy perturbation methods. However, under some limitations, MGDMP can be successfully applied to solve such type of RDEs.

Brain–computer interface (BCI) is the current trend in technology expansion as it provides an easy interface between human brain and machine. The demand for BCI-based applications is growing tremendously, and efforts are in progress to deploy BCI devices for real-world applications. One of the widely known applications of BCI technology is rehabilitation in which BCI devices can provide various types of assistance to specially abled persons. In this paper, the effect of hand actions on objects is analyzed for motor-related mental task. The proposed approach analyzes electroencephalogram (EEG)-based brain activity which was captured for images shown with different gripping actions on objects. The EEG recordings are first pre-processed, followed by extraction of epochs and frequency bands using discrete wavelet transform; afterward, feature extraction followed by training and classification steps is performed for classifying the grip action into congruent (correct) and incongruent (incorrect) grip categories. The proposed work makes use of average power and relative wavelet energy as discriminating features which are then fed to train an artificial neural network for automatically classifying the incoming EEG patterns into correct or incorrect object hand grips. The performance evaluation of the proposed system is done on real EEG dataset obtained from 14 subjects. Experimental results have shown an accuracy of 75%. Also, to evaluate the effectiveness of our work, a comparison of our work with other state-of-the-art works reported by different authors is presented at the end. The results show the effectiveness of proposed approach and suggest further that the system can be used to analyze and train subjects having motor-related disabilities for perceiving correct or incorrect hand grips on objects.

This paper presents, a thorough evaluation of popular deep learning models to analyze and classify electroencephalogram (EEG) data for characterizing human affective states for video content tagging and retrieval. We use two pre-trained convolutional neural network (CNN) models AlexNet and GoogLeNet, and a Long Short Term Memory (LSTM) model to classify EEG data into appropriate affect categories using trans-domain learning. The purpose behind the use of pre-trained networks or trans-domain learning is twofold – to establish the versatility of pre-trained networks by testing their ability to classify EEG data for emotion recognition and the other is to reduce over cost of computation while training the networks. Our work tries to establish the answer of a simple question: Are pre-trained deep models versatile enough for classifying not only similar type of problems but are also effective for classifying problems pertaining to completely different domains? Also, using pre-trained models saves considerable computation time required for training a new model from scratch and fine tuning it. We use DEAP dataset for training and evaluation of these networks over a single modality ‘valence’ to simplify the comparison among these networks. Experiments are carried out by training the networks on EEG recordings obtained from single as well as multiple subjects to show the effects of subject-specific and generalized data on classification accuracy. Experimental results suggest the superiority of GoogLeNet for individual subject data while AlexNet outperforms other networks and has shown its capability of generalizing well. We compare the performance of these networks with state-of-art classifiers handcrafted by other authors for classifying EEG data and find that the performance of pre-trained CNNs used in our work are comparable or even better than the other handcrafted classifiers used by many authors.

In this paper, we introduced deformable fractional derivative and fractional integral. Just like ordinary derivative, the deformable derivative is also defined using limit approach but with respect to a parameter, which vary over unit interval. Some basic properties are also discussed. As applications, we apply the D’Alambert approach to the deformable fractional differential equation of the type: \(D^{\alpha }D^{\alpha }y+PD^{\alpha }y+Qy=0\), where P and Q are \(\alpha \)-differentiable functions and representation of the solution of deformable fractional equation by using Mittag-Leffler function is also discussed.

In this paper, DCT (discrete cosine transform) with matrix technique for a fractional-order Riesz differentiator is presented. First, review the fractional-order Riesz differentiator definition. Next, transfer matrix of fractional-order Riesz differentiator is obtained by DCT interpolation method. Finally, some computative problems show the importance of DCT interpolation method are discussed.

The exact solution of every differential solution dose not exists though it is of integer order, and then there is need to turn toward the numerical solution or series solution. The same is the case with fractional order differential equation, likewise differential equations; series solutions also exist for the fractional ordered differential equations. This chapter studies the fractional power series solution of Laguerre and Chebyshev conformable differential equations using Katugampola's derivative which is the generalized form of conformable derivative. Existence and uniqueness theorem of linear conformable fractional differential equations around a point has been given and the existence of fractional power series solutions about a singular point of sequential conformable differential equation have been presented. The findings of the whole work indicate that the results of the fractional case conform with the results of the ordinary case.

The Sound source location estimation is usually done using an array of microphones. This kind of arrangement provides an environment where the phase or attenuation of the received signals from various microphones is known, which helps in sound sources localization. In this paper by using a microphone in the room, the room transfer function’s properties are exploiting for localization of sources of sound. Microphone’s position and shape of the room are assumed to be known. Sensing matrix’s limitations and design guidelines are given. On the basis of sparsity, implementation is done in a room in terms of voxels which are immersed by a source. Furthermore, solution’s exciting thing is that the sound source localization is provided with horizontal plane and 3D coordinates in a room.

This paper presents an ensemble of convolutional neural networks (CNNs) and support vector machine (SVM) for classifying emotions from electroencephalogram (EEG) patterns. We used popular deep learning models for feature extraction and a support vector machine classifier is employed to classify the EEG patterns into suitable emotion classes. The main contribution of this work is to investigate on the following points: creating an ensemble of pre-trained deep learning networks with support vector machine classifier (SVM) for classifying emotional states of person for single and multiple emotional attributes. Finding out the best ensemble network, extracting suitable layer and robust features to improve the classification accuracy of support vector machine and finally to compare the performance of ensemble of networks with stand-alone deep learning networks. Two popular convolutional neural networks are used for experiments: Alex Net and GoogLeNet. All experiments are carried out on database for emotion analysis using physiological signals (DEAP). A thorough analysis of experimental results revealed that classification accuracy of 87.5% is achieved by ensemble of Alex Net and SVM for single attribute (valance) classification while for two attributes (arousal and valance) the accuracy achieved is 62.5%. Similarly, accuracy of 100% and 62.5% are achieved for single and two attributes classification respectively using ensemble of GoogLeNet and SVM.

In this paper, we proposed a new analytical hybrid methods for the solution of conformable fractional differential equations (CFDE), which are based on the recently proposed conformable fractional derivative (CFD) in R. Khalil, M. Al Horani, A. Yusuf and M. Sababhed, A New definition of fractional derivative, J. Comput. Appl. 264 (2014). Moreover, we use the method of variation of parameters and reduction of order based on CFD, for the CFDE. Furthermore, to show the efficiency of the proposed analytical hybrid method, some examples are also presented.

Automated analysis of cervical cancer images is considered as an attractive research in the biological fields. Due to the intensive advances in the digital technology and the light microscopy, the cellular imaging requires the continuous growing importance. Sophisticated methods are adopted to isolate the nuclei from the cytoplasm based on the boundary estimation and the analysis of intensity of blue cells to improve the abnormality prediction. The cell-based segmentation evolved in research studies assures the automatic assistance with an assumption of a single cell. Previous work [21] concentrates on the segmentation of abnormal region on single-cell images. This paper extends that work into the multi-cell images with the addition of geometrical features. The complex cell structure, poor contrast, and overlapping affect the cell segmentation performance. This paper enhances the performance of single-cell segmentation with the integrated feature vectors of geometrical (area, cell size, cell intensity and the maximum intensity) and Gray level Co-occurrence Matrix (GLCM) to improve the abnormality level prediction.. Initially, the Neighborhood Concentric Filtering (NCF) is applied on the input slides to remove the noise present in the image and enhance the intensity level. Then, the initial level cluster formation and masking are performed on the noise-free image. The Optimal Weight Updating with the Multi-Level set (OWU-ML) estimates the Region of Interest (ROI) and segments the blue cell and cytoplasm. The clear analysis of blue cell indicates the exact classification of abnormal levels in the images. The combination of geometrical and GLCM extracts the texture pattern features of the blue cell, cytoplasm and the nucleus portions in the form of angle variations. Finally, the Neural Network-based RVM classifier predicts the classes of (normal and abnormal) cervical images. The integration of novel methods such as OWU-ML segmentation, GLCM + geometrical feature extraction and NN-RVM classification improves the abnormal prediction performance and assures the suitability in multi-cell cervical image handling in biological applications.

Intravascular ultrasound (IVUS) is a catheter-based imaging method used in the study of atherosclerotic disease. IVUS produces cross-sectional images of the blood vessels that enable quantitative assessment of the plaque. Automatic segmentation of the anatomical structures in the IVUS image is a really challenging task due to the presence of noise and catheter artifacts. Hence, this paper presents an efficient self-organizing map (SOM) and expectation-maximization (EM)-based approach for the segmentation of cross-sectional view of the IVUS blood vessel image. In our proposed work, the directional filtering is used to improve the signal to noise ratio of the blood vessel image. The Hough transform is used for predicting the circle in the image. Segmentation of the image is performed using the SOM and EM algorithm. After the segmentation process, extraction of the common pixels is performed. Gray-level co-occurrence matrix is applied for extracting features from the image. Fuzzy-relevance vector machine based classification of the image is performed. From the comparison results, it is clearly observed that the proposed approach is highly efficient than the existing techniques.

The computer vision field deals with the problem of understanding the scene or features in images of real world with the help of image processing and pattern recognition techniques. The main complication in this task is that the objects present in the images may have different appearances to the camera due to illumination effects, camera position, shadows, types of camera, etc. Nevertheless, with the advancement of technologies, today computer vision has provided reliable methods for various tasks like object classification, action recognition, autonomous driving, scene analysis, highlights extraction in videos and many more. But the problem of automatic qualifying is that how well people perform these actions has been largely unexplored. Human visual system and cognition can outperform the performance of computer vision algorithms. The objective of this paper is to highlight the state of the art of various psychological views of human visual perception in computer vision methods that have been found to operate well and that led up to the above-mentioned capabilities.

Action on any objects provides perceptual information about the environment. There is substantial evidence that human visual system responds to action possibilities in an image as perceiving any one’s action stimulates human motor system. However very limited studies have been done to analyze the effect of object affordance during action perception and execution. To study the effect of object affordance on human perception, in this paper we have analyzed the human brain signals using EEG based oscillatory activity of brain. EEG responses corresponding to images of objects shown with correct, incorrect and without grips are examined. Exploration of different gripping effects has been done by extracting Alpha and Beta frequency bands using Discrete Wavelet Transform based band extraction method, then baseline normalized power of Alpha and Beta frequency bands at 24 positions of motor area of left and right side of brain are examined. The result shows that twelve pooled electrodes at central and central parietal region provides a clear discrimination among the three gripping cases in terms of calculated power. The presented research explores new applicabilities of object affordance to develop a variety of Brain Computer Interface (BCI) based devices and to improve motor imagery ability among motor disorder related patients.

Image segmentation is a complex and essential task used in many computer vision applications. The problem of image segmentation can essentially be formulated as a grouping problem which in its simplest form tries to group the pixels of image into distinguished regions of interest so that further processing of the extracted regions can be achieved. This work proposes an image segmentation model which is inspired by the findings in cognitive psychology theories to divide the image into separate coherent regions. The proposed work tries to correlate between human and machine cognition by studying the segmentation process under the light of psychology of human vision.

Training of two or more tasks over shared representation involves in the Multi-task learning (MTL). In audio-visual automatic speech recognition, multi task learning is applied in the present work. Primary task of MTL is to learn mapping between frame labels obtained from acoustic GMM/HMM model and audio-visual fused features. An auxiliary task which maps frame labels obtained from a visual GMM/HMM model to visual features is combined with frame labels obtained from acoustic model and audio-visual fused features. Results of a base-line hybrid DNN-HMM AVASR model are compared with MTL model which is tested at various levels of babble noise. Results of this paper indicate that MTL is useful at higher level of noise. Comparison with the base-line model, at −3 SNR dB approximate 7%o relative improvement in WER is reported.

For i-vector model, normalization approach is Probabilistic linear discriminant analysis and has a significant performance for verification of speaker. However it requires a huge development data which cost a lot in many cases. Unsupervised adaption method is a possible approach, which use unlabeled data to adapt PLDA scattering matrices to the target domain. In this paper, ‘local training’ approach is adapted to train the PLDA model. In this approach local labels discriminate the speaker within a single conversion and therefore easy to obtain as compare to normal ‘global labels’. Proposed approach can deliver better performance and improvement, particularly with limited globally-labeled data.

On the basis of the evaluation of local properties of the data many nonlinear techniques have been suggested the field of computer vision. The application of the dimensionality reduction covers many fields like medical, geographical, simulation and many more. I have studied MDS, LLE and LTSA. Overall, the users are allowed to access the search-tools in linear system. A review and systematic comparison of all the existing techniques has been presented in this paper. The outputs have been explained through identification of current non-linear techniques, and suggestions pertaining to the way the performance of nonlinear dimensionality reduction techniques can be improved. The Purpose of this idea is based on the to implement it in manifold fields by analyzing the result of face detector and recognizer for multiple people in real time with Principal Component analysis on eigen face. According to the most recent research, some issues are confronted in the security at public places. The efficiency and accuracy of these problems can be improved with the range and intricacy of camera networks are booming and the audited surroundings have become more and more entangled and crowded. How these emerging challenges are faced is discussed in the paper.

In this paper, we introduce an application of recently proposed deformable derivative which is equivalent to ordinary derivative in the sense that one implies other. The deformable derivative is defined using limit approach as ordinary derivative. Thus it could also be regarded as fractional derivative. The simple nature of this definition allows us for the extension of some classical theorems in calculus like the Rolles, Mean Value and Extended Mean Value theorems. As a theoritical application some fractional differentiable equations are solved.

In this era of rapid technical developments in the field of sound source localisation, there is an urgent need to integrate these individual technologies to create an accurate and more power localiser. When the technique volumetric steered response power (V-SRP) is compared to steered response power (SRP), the former achieves a significant lower computational complexity. This is achieved without much decrease in the accuracy of location estimation. By appending a fine search stem to V-SRP, a refined version (RV-SRP) is achieved with improved compromise of complexity and accuracy. Experiments are performed using both the simulated as well as real scenario data to demonstrate the decrease in the computational cost. Further, in the experiment, the global positioning system (GPS) provides the accurate location coordinates of the source. It is to report newly developed system and methodologies which will be able to find approximate azimuth and elevation angle of sound source with integration of TDE-ILD-HRTF along with geo-positioning visualisations to help humanity to achieve its goal towards peace and harmony.

This paper proposes an Angular Texture Pattern (ATP)-Multi-Level Set Model (MLSM)-based retinal image segmentation approach. The location of Optical Disk (OD) is estimated by initially collecting the blood vessel region from the retinal image. Based on the identification of OD location, the bright pixel values are estimated to provide the boundary detail of OD. From this boundary detail, the Region of Interest (ROI) such as Hard Exudates (HE) is obtained in the binary form, to enable contour formation for the OD and HE. Then, the cup-to-disk ratio of the OD is calculated, and the number of HEs is counted. The severity level of the DR and Glaucoma is determined based on the cup-to-disk ratio of the OD and HE count value. The proposed algorithm is tested by using the retinal images of the DIARETDB1 and MESSIDOR database. The proposed approach achieves better performance than the existing OD segmentation methodologies.

The sole purpose of this study is to propose a new concept to fractional derivative as Ujlayan-Dixit derivative in the classical sense using limit approach. The parameter α of the derivative is confined within a closed unit interval. It is easy to apply and generates a convex combination of function and its derivative. We observe its geometrical behavior and see, how it is different from the previous definitions of the concerned derivatives. It works as an improved form of Conformable fractional derivative and fulfill the meaningful gaps. Some basic properties of calculus like continuity, differentiability, Mean Value theorem, Comparison theorem and some numerical have discussed. Publisher's Version