Search

Coronavirus is a fatal disease that affects mammals and birds. Usually, this virus spreads in humans through aerial precipitation of any fluid secreted from the infected entity's body part. This type of virus is fatal than other unpremeditated viruses. Meanwhile, another class of coronavirus was developed in December 2019, named Novel Coronavirus (2019-nCoV), first seen in Wuhan, China. From January 23, 2020, the number of affected individuals from this virus rapidly increased in Wuhan and other countries. This research proposes a system for classifying and analyzing the predictions obtained from symptoms of this virus. The proposed system aims to determine those attributes that help in the early detection of Coronavirus Disease (COVID-19) using the Adaptive Neuro-Fuzzy Inference System (ANFIS). This work computes the accuracy of different machine learning classifiers and selects the best classifier for COVID-19 detection based on comparative analysis. ANFIS is used to model and control ill-defined and uncertain systems to predict this globally spread disease's risk factor. COVID-19 dataset is classified using Support Vector Machine (SVM) because it achieved the highest accuracy of 100% among all classifiers. Furthermore, the ANFIS model is implemented on this classified dataset, which results in an 80% risk prediction for COVID-19.

Graphical abstract Most drugs are no longer produced in their own countries by the pharmaceutical companies, but by contract manufacturers or at manufacturing sites in countries that can produce more cheaply. This not only makes it difficult to trace them back but also leaves room for criminal organizations to fake them unnoticed. For these reasons, it is becoming increasingly difficult to determine the exact origin of drugs. The goal of this work was to investigate how exactly this is possible by using different spectroscopic methods like nuclear magnetic resonance and near- and mid-infrared spectroscopy in combination with multivariate data analysis. As an example, 56 out of 64 different paracetamol preparations, collected from 19 countries around the world, were chosen to investigate whether it is possible to determine the pharmaceutical company, manufacturing site, or country of origin. By means of suitable pre-processing of the spectra and the different information contained in each method, principal component analysis was able to evaluate manufacturing relationships between individual companies and to differentiate between production sites or formulations. Linear discriminant analysis showed different results depending on the spectral method and purpose. For all spectroscopic methods, it was found that the classification of the preparations to their manufacturer achieves better results than the classification to their pharmaceutical company. The best results were obtained with nuclear magnetic resonance and near-infrared data, with 94.6%/99.6% and 98.7/100% of the spectra of the preparations correctly assigned to their pharmaceutical company or manufacturer. Supplementary Information The online version contains supplementary material available at 10.1007/s00216-021-03249-z.

Potamotrygon motoro has been shown to use vision to orient in a laboratory setting and has been successfully trained in cognitive behavioral studies using visual stimuli. This study explores P. motoro’s visual discrimination abilities in the context of two-alternative forced-choice experiments, with a focus on shape and contrast, stimulus orientation, and visual resolution. Results support that stingrays are able to discriminate stimulus-presence and -absence, overall stimulus contrasts, two forms, horizontal from vertical stimulus orientations, and different colors that also vary in brightness. Stingrays tested in visual resolution experiments demonstrated a range of visual acuities from

SVM is a new general learning method based on the statistic learning system which can be used as an effective means to process small sample, nonlinear and high dimensional pattern recognition. This paper did research on the learning algorithm of support vector machine, extracted characteristic data of banknote which is on account of PCA according to the characteristics of the support vector machine (SVM), and proposed to put support vector machine (SVM) into banknotes denomination recognition by combining SMO training algorithm with one-to-many multi-value classification algorithm. Besides, this article used genetic algorithm in parameters optimization such as the punishment coefficient C of soft margin SVM and the width parameter of Gaussian kernel function. The ultimate purpose is to recognize the denomination of banknote efficiently and accurately. The experimental results verified that this kind of recognition method increases the recognition accuracy up to 90% or more.

Strategies used in artificial grammar learning can shed light into the abilities of different species to extract regularities from the environment. In the A(X)nB rule, A and B items are linked but assigned to different positional categories and separated by distractor items. Open questions are how widespread is the ability to extract positional regularities from A(X)nB patterns, which strategies are used to encode positional regularities and whether individuals exhibit preferences for absolute or relative position encoding. We used visual arrays to investigate whether cotton-top tamarins (Saguinus oedipus) can learn this rule and which strategies they use. After training on a subset of exemplars, half of the tested monkeys successfully generalized to novel combinations. These tamarins discriminated between categories of tokens with different properties (A, B, X) and detected a positional relationship between non-adjacent items even in the presence of novel distractors. Generalization, though, was incomplete, since we observed a failure with items that during training had always been presented in reinforced arrays. The pattern of errors revealed that successful subjects used visual similarity with training stimuli to solve the task, and that tamarins extracted the relative position of As and Bs rather than their absolute position, similarly to what observed in other species. Relative position encoding appears to be the default strategy in different tasks and taxa.

In this paper we describe a machine vision approach to recognize the denomination classes of the Mexican paper currency by extracting their color features. A banknote's color is characterized by summing all the color vectors of the image's pixels to obtain a resultant vector, the banknote's denomination is classified by knowing the orientation of the resulting vector within the RGB space. In order to obtain a more precise characterization of paper currency, the less discriminative colors of each denomination are eliminated from the images; the color selection is applied in the RGB and HSV spaces, separately. Experimental results with the current Mexican banknotes are presented.

To increase the reliability of currency classification, a classification method using neural networks with multi-pattern vectors is proposed in this paper. The data space of samples are divided into three blocks, then the latter are further divided into four sub-pattern vectors, and kernel principal component analysis is applied to extract features and assemble feature vectors to train LVQ neural network classifier. We draw the conclusion by testing new fifth edition RMB including four kinds of inputting directions of 1 Yuan, 5 Yuan, 10 Yuan and 20 Yuan RMB, up to 800 samples that PCA can compress data and decrease dimension of input vectors, extract the feature vectors effectively, thus the high-level reliability can be achieved by using the LVQ network classifier.

This paper is concerned with a multi-resolution tool for screening paper formation variations in various frequency regions on production line. A paper web is illuminated by two red diode lasers and the reflected light recorded as two time series of high resolution measurements constitute the input signal to the papermaking process monitoring system. The time series are divided into blocks and each block is analyzed separately. The task is treated as kernel based novelty detection applied to a multi-resolution time series representation obtained from the band-pass filtering of the Fourier power spectrum of the series. The frequency content of each frequency region is characterized by a feature vector, which is transformed using the canonical correlation analysis and then categorized into the inlier or outlier class by the novelty detector. The ratio of outlying data points, significantly exceeding the predetermined value, indicates abnormalities in the paper formation. The tools developed are used for online paper formation monitoring in a paper mill.

This paper is concerned with a multi-resolution tool for analysis of a time series aiming to detect abnormalities in various frequency regions. The task is treated as a kernel based novelty detection applied to a multi-level time series representation obtained from the discrete wavelet transform. Having a priori knowledge that the abnormalities manifest themselves in several frequency regions, a committee of detectors utilizing data dependent aggregation weights is build by combining outputs of detectors operating in those regions.

The main topic of the article is to give a detailed description of the new and promising quadtree decomposition texture analysis method used for paper formation determination. Paper formation or configuration of fibers, fines and fillers in the two-dimensional spatial xy-domain of the paper is a very important property and image analysis application for the paper industry. The basis of the method is the successive quadtree decomposition process resulting in a two-dimensional block partitioning of the formation structure image analysed. In this context the blocks represents a unit of variation, and the size of a quadtree block is controlled by a set of different parameters. In addition to the primary features detected by the algorithm, characterization of a large set of secondary features is performed, including gradient analysis and spatial distribution analysis.

This paper presents the experimental aspects of a robotic system used for detection, sorting and grading of paper objects from visual features. The experimental system is used as an automated paper recycling system. The system uses simplified stereo vision to compute position estimate of unknown objects. A vector of geometrical and textural features is used for sorting the paper, according to its grade. The grading process uses additional contact sensors for refining grading from visual features. Supervised learning is used for training the decision system to separate paper from other objects. The experimental aspects of the training process are presented in detail. A special vacuum gripper was designed for gripping paper objects detected by the vision system. The experimental results prove the efficiency of the proposed techniques.

A knowledge-based approach to retrieve medical images by feature and content with spatial and temporal constructs is developed. Selected objects of interest in a medical image (e.g. x-ray, MR image) are segmented, and contours are generated from these objects. Features (e.g. shape, size, texture) and content (e.g. spatial relationships among objects) are extracted and stored in a feature and content database. Knowledge about image features can be expressed as a hierarchical structure called a Type Abstraction Hierarchy (TAH). The high-level nodes in the TAH represent more general concepts than low-level nodes. Thus, traversing along TAH nodes allows approximate matching by feature and content if an exact match is not available. TAHs can be generated automatically by clustering algorithms based on feature values in the databases and hence are scalable to large collections of image features. Further, since TAHs are generated based on user classes and applications, they are context- and user-sensitive.

© The Author(s) 2017. The influence of an aortic aneurysm on blood flow waveforms is well established, but how to exploit this link for diagnostic purposes still remains challenging. This work uses a combination of experimental and computational modelling to study how aneurysms of various size affect the waveforms. Experimental studies are carried out on fusiform-type aneurysm models, and a comparison of results with those from a one-dimensional fluid–structure interaction model shows close agreement. Further mathematical analysis of these results allows the definition of several indicators that characterize the impact of an aneurysm on waveforms. These indicators are then further studied in a computational model of a systemic blood flow network. This demonstrates the methods’ ability to detect the location and severity of an aortic aneurysm through the analysis of flow waveforms in clinically accessible locations. Therefore, the proposed methodology shows a high potential for non-invasive aneurysm detectors/monitors.

Face verification technology is widely used in the fields of public safety, e-commerce and so on. Due its characteristic of insensitive to the varied illumination, a new method about face verification with illumination invariant is presented in this paper based on gabor wavelet. First, ATICR method is used to do light preprocessing on images. Second, certain gabor wavelet filters, which are selected on the experiment inducing different gagor wavelet filter has not the same effect in verification, are used to extract feature of the image, of which the dimension in succession is reduced by Principal Component Analysis. At last, SVM classifiers are modeled on the data with reduced dimension. The experiment results in IFACE database and NIRFACE database indicate the algorithm named “Selected Paralleled Gabor Method” can achieves higher verification performance and better adaptability to the variable illumination.

In this short paper we shall consider the Kernel Fisher Discriminant Analysis (KFDA) and extend the idea of Linear Discriminant, Analysis (LDA) to nonlinear feature space. We shall present a new method of choosing the optimal kernel function and its effect on the KDA classifier using information-theoretic complexity measure.

The world is still under the threat of different strains of the coronavirus and the pandemic situation is far from over. The method, that is widely used for the detection of COVID-19 is Reverse Transcription Polymerase chain reaction (RT-PCR), which is a time-consuming method and is prone to manual errors, and has poor precision. Although many nations across the globe have begun the mass immunization procedure, the COVID-19 vaccine will take a long time to reach everyone. The application of artificial intelligence (AI) and computer-aided diagnosis (CAD) has been used in the domain of medical imaging for a long period. It is quite evident that the use of CAD in the detection of COVID-19 is inevitable. The main objective of this paper is to use convolutional neural network (CNN) and a novel feature selection technique to analyze Chest X-Ray (CXR) images for the detection of COVID-19. We propose a novel two-tier feature selection method, which increases the accuracy of the overall classification model used for screening COVID-19 CXRs. Filter feature selection models are often more effective than wrapper methods as wrapper methods tend to be computationally more expensive and are not useful for large datasets dealing with a large number of features. However, most filter methods do not take into consideration how a group of features would work together, rather they just look at the features individually and decide on a score. We have used approximate Shapley value, a concept of Coalition game theory, to deal with this problem. Further, in the case of a large dataset, it is important to work with shorter embeddings of the features. We have used CUR decomposition and Nystrom sampling to further reduce the feature space. To check the efficacy of this two-tier feature selection method, we have applied it to the features extracted by three standard deep learning models, namely VGG16, Xception and InceptionV3, where the features have been extracted from the CXR images of COVID-19 datasets and we have found that the selection procedure works quite well for the features extracted by Xception and InceptionV3. The source code of this work is available at https://github.com/subhankar01/covidfs-aihc.

Current available supervised classifiers cannot generalize across various domains due to distribution mismatch among them. Domain adaptation and transfer learning algorithms are proposed to tackle domain shift problem that originates from different data collection conditions. In this paper, we propose a transfer learning framework called iterative joint classifier and domain adaptation for visual transfer learning (ICDAV), which utilizes the balanced maximum mean discrepancy to better transfer knowledge across domains. Also, for learning a robust classifier against domain shift, a set of graph manifold regularizer and modified joint probability maximum mean discrepancy are simultaneously exploited to capture the domain structures and adapt the distribution of projected samples during the model learning process. Variety of experiments on several public datasets indicates that our approach achieves remarkable performance on visual domain adaptation and transfer learning tasks.

Image classification is getting more attention in the area of computer vision. During the past few years, a lot of research has been done on image classification using classical machine learning and deep learning techniques. Presently, deep learning-based techniques have given stupendous results. The performance of a classification system depends on the quality of features extracted from an image. The better is the quality of extracted features, the more the accuracy will be. Although, numerous deep learning-based methods have shown enormous performance in image classification, still due to various challenges deep learning methods are not able to extract all the important information from the image. This results in a reduction in overall classification accuracy. The goal of the present research is to improve the image classification performance by combining the deep features extracted using popular deep convolutional neural network, VGG19, and various handcrafted feature extraction methods, i.e., SIFT, SURF, ORB, and Shi-Tomasi corner detector algorithm. Further, the extracted features from these methods are classified using various machine learning classification methods, i.e., Gaussian Naive Bayes, Decision Tree, Random Forest, and eXtreme Gradient Boosting (XGBClassifier) classifier. The experiment is carried out on a benchmark dataset Caltech-101. The experimental results indicate that Random Forest using the combined features give 93.73% accuracy and outperforms other classifiers and methods proposed by other authors. The paper concludes that a single feature extractor whether shallow or deep is not enough to achieve satisfactory results. So, a combined approach using deep learning features and traditional handcrafted features is better for image classification.

The presentation of the COVID19 has endangered several million lives worldwide causing thousands of deaths every day. Evolution of COVID19 as a pandemic calls for automated solutions for initial screening and treatment management. In addition to the thermal scanning mechanisms, findings from chest X-ray imaging examinations are reliable predictors in COVID19 detection, long-term monitoring and severity evaluation. This paper presents a novel deep transfer learning based framework for COVID19 detection and segmentation of infections from chest X-ray images. It is realized as a two-stage cascaded framework with classifier and segmentation subnetwork models. The classifier is modeled as a fine-tuned residual SqueezeNet network, and the segmentation network is implemented as a fine-tuned SegNet semantic segmentation network. The segmentation task is enhanced with a bioinspired Gaussian Mixture Model-based super pixel segmentation. This framework is trained and tested with two public datasets for binary and multiclass classifications and infection segmentation. It achieves accuracies of 99.69% and 99.48% for binary and three class classifications, and a mean accuracy of 83.437% for segmentation. Experimental results and comparative evaluations demonstrate the superiority of this unified model and signify potential extensions for biomarker definition and severity quantization.

Coronavirus (COVID-19) is an epidemic that is rapidly spreading and causing a severe healthcare crisis resulting in more than 40 million confirmed cases across the globe. There are many intensive studies on AI-based technique, which is time consuming and troublesome by considering heavyweight models in terms of more training parameters and memory cost, which leads to higher time complexity. To improve diagnosis, this paper is aimed to design and establish a unique lightweight deep learning-based approach to perform multi-class classification (normal, COVID-19, and pneumonia) and binary class classification (normal and COVID-19) on X-ray radiographs of chest. This proposed CNN scheme includes the combination of three CBR blocks (convolutional batch normalization ReLu) with learnable parameters and one global average pooling (GP) layer and fully connected layer. The overall accuracy of the proposed model achieved 98.33% and finally compared with the pre-trained transfer learning model (DenseNet-121, ResNet-101, VGG-19, and XceptionNet) and recent state-of-the-art model. For validation of the proposed model, several parameters are considered such as learning rate, batch size, number of epochs, and different optimizers. Apart from this, several other performance measures like tenfold cross-validation, confusion matrix, evaluation metrics, sarea under the receiver operating characteristics, kappa score and Mathew's correlation, and Grad-CAM heat map have been used to assess the efficacy of the proposed model. The outcome of this proposed model is more robust, and it may be useful for radiologists for faster diagnostics of COVID-19.

Different respiratory infections cause abnormal symptoms in lung parenchyma that show in chest computed tomography. Since December 2019, the SARS-COV-2 virus, which is the causative agent of COVID-19, has invaded the world causing high numbers of infections and deaths. The infection with SARS-COV-2 virus shows an abnormality in lung parenchyma that can be effectively detected using Computed Tomography (CT) imaging. In this paper, a novel computer aided framework (COV-CAF) is proposed for classifying the severity degree of the infection from 3D Chest Volumes. COV-CAF fuses traditional and deep learning approaches. The proposed COV-CAF consists of two phases: the preparatory phase and the feature analysis and classification phase. The preparatory phase handles 3D-CT volumes and presents an effective cut choice strategy for choosing informative CT slices. The feature analysis and classification phase incorporate fuzzy clustering for automatic Region of Interest (RoI) segmentation and feature fusion. In feature fusion, automatic features are extracted from a newly introduced Convolution Neural Network (Norm-VGG16) and are fused with spatial hand-crafted features extracted from segmented RoI. Experiments are conducted on MosMedData: Chest CT Scans with COVID-19 Related Findings with COVID-19 severity classes and SARS-COV-2 CT-Scan benchmark datasets. The proposed COV-CAF achieved remarkable results on both datasets. On MosMedData dataset, it achieved an overall accuracy of 97.76% and average sensitivity of 96.73%, while on SARS-COV-2 CT-Scan dataset it achieves an overall accuracy and sensitivity 97.59% and 98.41% respectively.

Xinjiang carpet patterns have flowery color, strong contrast, and the overall harmony, how to design a carpet patterns with Xinjiang style feature is a challenging problem. This paper puts forward a method based on interactive choice mechanism and establishment color constraint rules in order to create Xinjiang carpet patterns. First, we can establish pattern design model through the medallion pattern, corner pattern and brink pattern; then according to the user’s choice, we can get sample pattern, and establish respectively dominant color matrix of samples and generated patterns; finally, through the color similarity to generate design constraints, the simulation experiment shows that method of the paper can innovatively design more inherited color style of Xinjiang carpet designs, and enrich the research methods of Xinjiang carpet style design.

This paper proposes a novel method, which is coined as ARBBPNN, for biometric-oriented face detection, based on autoregressive model with Bayes backpropagation neural network (BBPNN). Firstly, the given input colour key face image is modelled to HSV and YCbCr models. A hybrid model, called HS–YCbCr, is formulated based on the HSV and YCbCr models. The submodel, H, is divided into various sliding windows of size, 3 × 3. The model parameters are estimated for the window using the BBPNN. Based on the model coefficients, autocorrelation coefficients (ACCs) are computed. An autocorrelation test tests the significance of the ACCs. If the ACC passes the test, then it is inferred that the small image region, viz. the window, represents the texture and it is treated as the texture feature. Otherwise, it is regarded as structure, which is treated as the shape feature. The texture and shape features are formulated as feature vectors (FV) separately, and they are combined into a single FV. This process is performed for all colour submodels. The FVs of the submodels are combined into a single holistic vector, which is treated as the FV of the key face image. The key FV has twenty feature elements. The similarity of the key and target face images is examined, based on the key and target FVs, by deploying multivariate parametric statistical tests. If the FVs of the key and target images pass the tests, then it is concluded that the key and target face images are the same; otherwise, they are regarded as different. The GT, FSW, Pointing’04, and BioID datasets are considered for the experiments. In addition to the above datasets, we have constructed a dataset with face images collected from Google, and many images captured through a digital camera. It is also subjected to the experiment. The obtained recognition results show that the proposed ARBBPNN method outperforms the existing methods.

Conventional visual and manual road crack detection method is labor-consuming, non-precise, dangerous, costly and also it can affect transportation. With crack being the main distress in the actual pavement surface, digital image processing has been widely applied to cracking recognition recently. This paper presents the preprocessing method, segmentation method, the locating method, and a novel convolutional neural network based pavement cracking recognition method in the area of image processing. This paper trains and tests aforementioned 5-layer convolutional neural network on the pavement crack dataset. The experimental result shows that this 5-layer convolutional neural network performs better than that classical conventional machine learning method. Actual pavement images are used to verify the performance of this method, and the results show that the surface crack could be identified correctly and automatically. The convolutional neural network can learn the features of crack well and sort these aircraft with a high classification accuracy.

In general, character recognition consists of four stages: image preprocessing, segmentation, feature extraction, and classification. Character segmentation is one of the most important and difficult tasks in character recognition. Incorrectly segmented characters are not likely to be correctly recognized. Touching characters, which always arises when handwritten characters are being segmented, makes the task even more difficult. Therefore, this paper emphasizes the interest to the segmentation of touching and overlapping characters. This paper proposes two new techniques which are shown to dramatically improve the segmentation accuracy. The first proposed technique emphasizes on converting a greyscale image to a binary image while the second proposed technique emphasizes on the process of character segmentation itself. In the proposed character segmentation process, the bounding box analysis is initially employed to segment the document image into images of isolated characters and images of touching characters. The thinning algorithm is applied to extract the skeleton of the touching characters. Next, the skeleton of the touching characters is separated into several pieces. Finally, the separated pieces of the touching characters are put back to reconstruct two isolated characters. The proposed algorithm achieves an accuracy of 89.26%.

With the increasing applications of computing systems, recognizing accurate and application oriented human expressions, is becoming a challenging topic. The face is a highly attractive biometric trait for expression recognition because of its physiological structure and location. In this paper we proposed two different subspace projection methods that are the extensions of basis subspace projection methods and applied them successfully for facial expression recognition. Our first proposal is an improved principal component analysis for facial expression recognition in frontal images by using an extension of eigenspaces and we term this as WR-PCA (region based and weighted principal component analysis). Secondly we proposed locally salient Independent component analysis(LS-ICA) method for facial expression analysis. These two methods are extensively discussed in the rest of the paper. Experiments with Cohn-kanade database show that these techniques achieves an accuracy rate of 93% when using LS-ICA and 91.81% when WR-PCA and 83.05% when using normal PCA. Our main contribution here is that by performing WR-PCA, which is an extension of typical PCA and first investigated by us, we achieve a nearly similar result as LS-ICA which is a very well established technique to identify partial distortion.

This paper proposes a biometric system based on features extracted from the thyroid tissue accessed through 2D ultrasound. Tissue echo-morphology, which accounts for the intensity (echogenicity), texture and structure has started to be used as a relevant parameter in a clinical setting. In this paper, features related to texture, morphology and tissue reflectivity are extracted from the ultrasound images and the most discriminant ones are selected as an input for a prototype biometric identification system. Several classifiers were tested, with the best results being achieved by a combination of classifiers (k-Nearest Neighbors, MAP and entropy distance). Using leave-one-out cross-validation method the identification rate was up to 94%. Features related to texture and echogenicity were tested individually with high identification rates up to 78% and 70%, respectively. This suggests that the acoustic impedance (reflectivity or echogenicity) of the tissue as well as texture are feasible parameters to discriminate between distinct subjects. This paper shows the effectiveness of the proposed classification, which can be used not only as a new biometric modality but also as a diagnostic tool.

This paper presents a method for machine component supervision with little to none prior knowledge of the machine, operating conditions and wear behavior. A hybrid approach based on unsupervised learning methods, consisting of an autoencoder network and clustering, to identify machine states and possible failure preceding anomalies is proposed. In order to cope with information sparsity, the model parameters of the unsupervised methods are derived automatically based on data distribution and a physical motivation. The approach was validated on a dataset of artificially introduced bearing faults. The gained clustering results show a general usability of the approach for condition monitoring with vibration data.

There exist a number of applications that make use of automatic facial expression synthesis and analysis, especially for interaction or communication between human and computers. This paper proposes a novel approach for facial expression synthesis that can generate realistic expressions for a new person with natural expression details. This approach is based on local geometry preserving between the input face image and the target expression image. In order to generate expressions with arbitrary intensity and mixed expression types, this paper also develops an expression analysis scheme based on Supervised Locality Preserving Projections (SLPP) that aligns different subjects and different intensities on a generalized expression manifold. Experimental results demonstrate the effectiveness of the proposed algorithm.

Conventional human abnormal behavior detection is mostly done in videos taken by visible-light cameras, and it is usually designed for a certain task. In order to solve the human abnormal behavior detection problem in general situation, this paper proposes a detection algorithm based on skeleton information entropy, by using the information from RGBD videos. In this paper, we assume that abnormal behavior is disordered. To sample the accurate features of human, we use RGBD cameras to get the skeleton information. Then, we analyze the information entropy of the angles of the skeleton, and find that the values of the information entropy are significantly higher in abnormal videos than in normal videos. The methods are tested in our database taken by Kinect in our lab and we present superior results whose recall is 92 % and precision is 95.83 %, and accuracy is 94 %.

Rapid development of unattended substation requires that computer image recognition technology should be applied to power system more imminently. This paper analyzes the advantages of applying fuzzy pattern recognition technique to electric online monitoring image color identification and classification, and proposes a new method. Firstly, the algorithm acquires all kinds of identification color center based on sample learning sets given by experts; then, it introduces fuzzy c-means (FCM) clustering method and connected graph traversal technique. Based on the identification of color membership of pixel’s corresponding color pattern and color’s no-mutation rules, this paper comprehensively analyzes the whole image to form each color pattern and obtain the color identification results of each area. This method is prevalently instructional to similar color automatic identification.

The problem of Text Classification TC has been studied for decades, and this problem is particularly interesting because the features are derived from syntactic or semantic indicators, while the classification, in and of itself, is based on statistical Pattern Recognition PR strategies. Thus, all the recorded TC schemes work using the fundamental paradigm that once the statistical features are inferred from the syntactic/semantic indicators, the classifiers themselves are the well-established ones such as the Bayesian, the Naive Bayesian, the SVM etc. and those that are neural or fuzzy. In this paper, we shall demonstrate that by virtue of the skewed distributions of the features, one could advantageously work with information latent in certain "non-central" quantiles i.e., those distant from the mean of the distributions. We, indeed, demonstrate that such classifiers exist and are attainable, and show that the design and implementation of such schemes work with the recently-introduced paradigm of Quantile Statistics QS-based classifiersThe foundational properties for CMQS for generic and some straightforward distributions were initially described in [17]. Their properties for uni-dimensional distributions of the exponential family are included in [9], and for multi-dimensional distributions in [18]. The authors of [17],i¾?[9] and [18] had initially proposed their results as being based on the Order-Statistics of the distributions. This was later corrected in [19], where they showed that their results were rather based on their Quantile Statistics.. These classifiers, referred to as Classification by Moments of Quantile Statistics CMQS, are essentially "Anti"-Bayesian in their modus operandi. To achieve our goal, in this paper we demonstrate the power and potential of CMQS to describe the very high-dimensional TC-related vector spaces in terms of a limited number of "outlier-based" statistics. Thereafter, the PR task in classification invokes the CMQS classifier for the underlying multi-class problem by using a linear number of pair-wise CMQS-based classifiers. By a rigorous testing on the standard 20-Newsgroups corpus we show that CMQS-based TC attains accuracy that is comparable to the best-reported classifiers. We also propose the potential of fusing the results of a CMQS-based methodology with those obtained from a more traditional scheme.

Image super-resolution focuses on achieving the high-resolution version of single or multiple low-resolution images. In this paper, a novel super-resolution approach based on morphological component analysis (MCA) and dictionary learning is proposed in this paper. The approach can recover each hierarchical structure well for the reconstructed image. It is integrated mainly by the dictionary learning step and high-resolution image reconstruction step. In the first step, the high-resolution and low-resolution dictionary pairs are trained based on MCA and sparse representation. In the second step, the high-resolution image is reconstructed by the fusion between the high-resolution cartoon part and texture part. The cartoon is acquired by MCA from the interpolated source image. The texture is recovered by the dictionary pairs. Experiments show that the desired super-resolution results can be achieved by the approach based on MCA and dictionary learning.

Feature extraction is one of the important characteristics used in classifying images. But the extracted features have big numbers and high dimension easily due to various type defects, complicated features and diverse methods of feature. Big numbers and high dimension of features are adverse for feature extraction. The effect of feature extraction decides the effect of image classification directly. According to these problems, experimental investigations are carried out on computer aiming at three typical surface defect images of cold steel strip, and this paper choose the gray features, textural features and Hu invariant moment features as the basis of classification finally. Experimental results demonstrated that features in this paper can be classification basis correctly.

This paper focuses on outlier detection from the perspective of classification. One-class support vector machine (OCSVM) is a widely applied and effective method of outlier detection. Unfortunately experiments show that the standard one-class SVM is easy to be influenced by the outliers contained in the training dataset. To cope with this problem, a robust OCSVM is presented in the paper. In consideration that the contribution yielded by the outlying instances and the normal data is different, a robust one-class SVM which assigns an adapting weight for every object in the training dataset was proposed in this paper. Experimental analysis shows the better performances of the proposed weighted method compared to the conventional one-class SVM on robustness.

This paper is focused on the acoustic events detection. Particularly two types of acoustic events (gun shot, breaking glass) were investigated. For any detection task the feature extraction methods play very important role. The feature extraction influences the recognition rate, therefore it is most important in any pattern recognition task. In this paper the impact of Mel-Frequency Cepstral Coefficients - MFCC and selected set of MPEG-7 low-level descriptors were examined. The best feature set contained MFCC and selected descriptors such as ASC, ASS, ASF. They were used to represent the sounds of acoustic events and background. We obtained the improvement of the detection rate using the mentioned set of features. In this task GMM classifiers are used to model the sound classes. This paper describes a basic aspect of our work.

This paper presents two classification systems for medical images based on association rule mining. The system we propose consists of: a pre-processing phase, a phase for mining the resulted transactional database, and a final phase to organize the resulted association rules in a classification model. The experimental results show that the method performs well, reaching over 80% in accuracy. Moreover, this paper illustrates how important the data cleaning phase is in building an accurate data mining architecture for image classification.

Foliage morphological features are important for plant recognition. However, the foliage shape generally presents big intra-class variations and small inter-class differences. This brings a great challenge to accurate plant foliage recognition. In this paper, we propose a deep residual squeeze-excitation network (R-SENet) for foliage recognition. Firstly, R-SENet learns and obtains the significance levels of each channel of the various convolutional layers in a residual block to recognition tasks via squeeze-excitation strategy. Then, the weights of each channel are rescaled by means of the significances to promote the relevant channels and inhibit non-important channels. Finally, we evaluate the proposed approach on the well-known Flavia dataset for foliage recognition. The experimental results indicate that our approach achieves more accurate average recognition rate (up to 97.86%) and more robustness to noise than other outstanding approaches.

Multiview object classification and recognition is of great importance in many applications such as computer vision and robotics by Kuo and Nevatia (Applications of computer vision (WACV), p. 18, 2009). This paper focuses on the specific case of multiview pedestrian image classification. The contributions of this paper are twofold. First, we collected a new multiview pedestrian dataset, which has been manually labeled with viewpoint, posture, and scene category tags. Second, a nonnegative least square (NNLS)-based multiview pedestrian image classification method is presented by Pang et al. (IEEE transactions on image processing, vol 20, pp. 1388–1400). Experimental results demonstrate that the proposed method is robust and effective.

There is a growing interest in subspace learning techniques for face recognition. This paper proposes a novel face recognition method based on MPCA with Gabor tensor representation. Although the Gabor face representation has achieved great success in face recognition, the excessive dimension of the data space often brings the algorithms into the curse of dimensionality dilemma. In this paper, we propose a 3rd-order Gabor tensor representation derived from a complete response set of 40 Gabor filters. Then MPCA (Multi-linear Principal Component Analysis) is applied to each Gabor tensor to extract three discriminative subspaces. The dimension reduction is done in such a way that most useful information is retained. The subspaces are finally integrated for classification. Experimental results on ORL database show promising results of the proposed method.

Cervical Cancer is one of the ubiquitous forms of cancer afflicting the female population worldwide. A Digital Colposcope is a self-illuminated powerful microscope which acquires the image of the affected cervix. The raw cervix image acquired by the colposcope is known as a cervigram. The raw cervigram is preprocessed by removing the specular reflections and then the region of interest is sought. Different novel segmentation algorithms which were proposed in our earlier papers are applied to the cervigram to detect the acetowhite tissues. In the current paper, we try to draw a correlation between Pap smear results, colposcopic findings, histopathologic biopsy results and our proposed algorithmic analysis. The significance of association is assessed by chi-square testing and the strength of association by kappa statistic. The association between impression and histology was significant, the strength of correlation was also significant. The overall kappa statistic for the concordance of the algorithm with the human experts was 0.809, while that among raters, excluding the algorithm was 0.804. The proposed algorithm had better agreement with the expert decision.

Convolutional neural networks (CNNs) have exhibited great potential in the field of image classification in the past few years. In this paper, we present a novel strategy named cross-level to improve the existing CNNs’ architecture in which different levels of feature representation in a network are merely connected in series. The basic idea of cross-level is to establish a convolutional layer between two nonadjacent levels, aiming to learn more sufficient feature representations. The proposed cross-level strategy can be naturally combined into a CNN without any change on its original architecture, which makes this strategy very practical and convenient. Three popular CNNs for image classification are employed to illustrate its implementation in detail. Experimental results on the dataset adopted by the ImageNet Large-Scale Visual Recognition Challenge (ILSVRC) verify the effectiveness of the proposed cross-level strategy on image classification. Furthermore, a new CNN with cross-level architecture is introduced in this paper to demonstrate the value of the proposed strategy in the future CNN design.

Facial image recognition has been widely used and implemented in many aspects of life such as, investigation area or security purposes. However, research in this area is still been done. Source images of this paper are taken from image library provided in description of the Collection of Facial Images, 2008. This paper explains how 35 faces in JPG format with dimension 180 x 180 pixels can be represented by only 3 x 3 DCT coefficients and can be recognized fully 100% by Radial Basis Function Network.

This paper addresses the problem of feature-based 3D reconstruction model for close-range objects. Since it is almost impossible to find pixel-to-pixel correspondences from 2D images by algorithms when the object is imaged on a close range, the selection of feature correspondences, as well as the number and distribution of them, play important roles in the reconstruction accuracy. Then, features on representative objects are analyzed and discussed. The impact of the number and distribution of feature correspondences is analyzed by reconstructing an object with standard cylinder shape by following the reconstruction model introduced in the paper. After that, three criteria are set to guide the selection of feature correspondences for more accurate 3D reconstruction. These criteria are finally applied to the human finger since it is a typical close-range object and different number and distribution of feature correspondences can be established automatically from its 2D fingerprints. The effectiveness of the setting criteria is demonstrated by comparing the accuracy of reconstructed finger shape based on different fingerprint feature correspondences with the corresponding 3D point cloud data obtained by structured light illumination (SLI) technique which is taken as a ground truth in the paper.

In order to enhance the recognition performance of the i-vector speaker recognition system under unpredicted noise environment, an improved local preserve projection which was used for reduce dimension to i-vector is proposed on this paper. First, the nonzero eigenvalue is rejected when we solve the optimized objective function, only using the eigenvalue the value of which is greater than zero. A mapping matrix is obtained by solving a generalized eigenvalue problem which can settle the singular value problem which occurred in the traditional local preserve projection algorithm. The experiment result shows that the recognition performance of the method proposed in this paper is improved under several kinds of noise environments.

This paper discusses possibilities of using ensembles of neural-networks-based classifiers in pattern recognition and classification. Attention is paid to systems that minimize demands on data preprocessing. Minimizing of requirements for preprocessing leads automatically to systems that are able to sufficiently classify the submitted data into predefined classes without knowledge of details of their significance. In our experiment, we try to increase diversity of classifiers by various filtering methods. The methods proposed in this paper come out from a technique called boosting, which is based on the principle of combining a large number of so-called weak classifiers into a strong classifier. All proposed improvements are experimentally verified.

In this paper, we focus on phase-only correlation-based im-age matching, which is often used for biometrics image retrieval. Al-though systems using phase-only correlation of images allow us to have image matching results of high quality, it takes long time to process phase-only correlation-based image matching especially when there are a large number of images to be processed in the matching. In this paper, we propose an approximate matching preprocessing for pruning images that do not have to be matched to a query image. An empirical study shows the superiority of our proposal.

Noninvasive assessing the physiological changes occurring inside the human brain is a challenging problem in biomedical engineering. These variations can be modeled as biomedical source signals that can be measured by several types of noninvasive brain imaging techniques such as electroencephalography (EEG). In this paper, after the perspective of linear blind source separation (BSS) model and characteristics of EEG are presented, the general and detailed definition of BSS model for EEG data analysis is given. Then based on the spatial structure and temporal or spectral information of the EEG signals, some state-of-the-art BSS techniques that can be used for analyzing EEG recordings are reviewed. A novel algorithm combining both high-order statistics and second-order statistics to achieve BSS for EEG is constructed. The paper concludes by discussing the influence of BSS for EEG research.

Clothes classification is a promising research topic. Due to the manually-designed features’ limitation, the existing algorithms have a problem of low accuracy in attributes classification. In this paper, we propose a new method to utilize convolutional deep learning for clothes classification. We firstly set up a new database by downloading the images of each category from Internet via related software and manual work, which divides clothes into 16 categories according to the common clothing style in the market. Then, the paper designs convolutional neural networks(CNNs) architecture and adaptively learns the feature representation of clothes from our constructed dataset. The experiment adopts Bag of Words (BOW), Histogram of Oriented Gradient (HOG)+ Support Vector Machine(SVM)and HSV (Hue, Saturation, Value)+SVM to test the new database and compares these methods with our CNNs model. The results demonstrate the superiority of our CNNs to the other algorithms.

Oil palm cultivation is one of the most important occupation in South East Asia. Since oil palm plantations cover the wide range of area, it is difficult to count the population of oil palm manually. This paper presents a new method to detect and identify oil palms over the plantation from aerial images regard- less of their sizes using features such as shape, size, and texture. The proposed method can handle the problem of identifying oil palms from an aerial image when oil palms are too close to each other which cause them to be detected as a single stand. The process consists of removing non-tree components from an image, distinguishing oil palms from other components, identifying individual oil palm, and counting the number of oil palms. In this paper, oil palm can be detected and distinguished from other components using ideal low-pass filter and normalized cross correlation. Then a proposed multi-scale clustering meth- od and erosion are used to identify individual oil palm from a bush. The method was evaluated on a set of 21 aerial images taken over oil palm plantations from different regions of Thailand by attaching a digital camera to a remote airplane. The experimental results reveal that our proposed method can detect most of the oil palms with the accuracy as high as 90%.