Search

Local Binary Patterns (LBP) have emerged as one of the most prominent and widely studied local texture descriptors. Truly a large number of LBP variants has been proposed, to the point that it can become overwhelming to grasp their respective strengths and weaknesses, and there is a need for a comprehensive study regarding the prominent LBP-related strategies. New types of descriptors based on multistage convolutional networks and deep learning have also emerged. In different papers the performance comparison of the proposed methods to earlier approaches is mainly done with some well-known texture datasets, with differing classifiers and testing protocols, and often not using the best sets of parameter values and multiple scales for the comparative methods. Very important aspects such as computational complexity and effects of poor image quality are often neglected. In this paper, we provide a systematic review of current LBP variants and propose a taxonomy to more clearly group the prominent alternatives. Merits and demerits of the various LBP features and their underlying connections are also analyzed. We perform a large scale performance evaluation for texture classification, empirically assessing forty texture features including thirty two recent most promising LBP variants and eight non-LBP descriptors based on deep convolutional networks on thirteen widely-used texture datasets. The experiments are designed to measure their robustness against different classification challenges, including changes in rotation, scale, illumination, viewpoint, number of classes, different types of image degradation, and computational complexity. The best overall performance is obtained for the Median Robust Extended Local Binary Pattern (MRELBP) feature. For textures with very large appearance variations, Fisher vector pooling of deep Convolutional Neural Networks is clearly the best, but at the cost of very high computational complexity. The sensitivity to image degradations and computational complexity are among the key problems for most of the methods considered. [ABSTRACT FROM AUTHOR]

We extend previously defined particle filters for lane detection by using a more general lane model supporting the use of two independent particle filters for detecting left and right lane borders separately, by combining multiple particles, traditionally used for identifying a winning particle in one image row, into one superparticle, and by using local linear regression for adjusting detected border points. The combination of multiple particles makes it possible to extend the traditional emphasis of particle-filter-based lane detectors (on identifying sequences of isolated border points) towards a local approximation of lane borders by polygonal or smooth curves further detailed in our local linear regression. The paper shows by experimental studies that results, obtained by the proposed novel lane detection procedure, improve compared to previously achieved particle-filter-based results especially for challenging lane detection situations. The presentation of several methods for comparative performance evaluation is another contribution of this paper. [ABSTRACT FROM AUTHOR]

The paper evaluates iterative semi-global matching (iSGM) and linear belief-propagation matching (linBPM), both using a census data-cost function, which are two of the currently top-ranked stereo matchers. The evaluation is on long real-world video sequences where disparity ground-truth is not available. The paper applies two alternative (or mutually supporting) techniques for performance evaluation: the previously known third-eye method, and a few new data measures on video sequences. The main contribution of the paper is on answering the questions, how to evaluate stereo matchers on long real-world sequences if disparity ground truth is not available, and how to compare evaluation measures relatively to each other. The two stereo matchers used are illustrating the discussed evaluation measures; they could be replaced by other matchers, but evaluation results for those two matchers are also of interest on its own, by illustrating correlations in the behavior of those two basically very different matchers (defined by dynamic programming or by belief propagation optimization, respectively) on data sequences recorded in different traffic situations. [ABSTRACT FROM AUTHOR]

Abstract: This paper presents a method for improved ensemble learning, by treating the optimization of an ensemble of classifiers as a compressed sensing problem. Ensemble learning methods improve the performance of a learned predictor by integrating a weighted combination of multiple predictive models. Ideally, the number of models needed in the ensemble should be minimized, while optimizing the weights associated with each included model. We solve this problem by treating it as an example of the compressed sensing problem, in which a sparse solution must be reconstructed from an under-determined linear system. Compressed sensing techniques are then employed to find an ensemble which is both small and effective. An additional contribution of this paper, is to present a new performance evaluation method (a new pairwise diversity measurement) called the roulette-wheel kappa-error. This method takes into account the different weightings of the classifiers, and also reduces the total number of pairs of classifiers needed in the kappa-error diagram, by selecting pairs through a roulette-wheel selection method according to the weightings of the classifiers. This approach can greatly improve the clarity and informativeness of the kappa-error diagram, especially when the number of classifiers in the ensemble is large. We use 25 different public data sets to evaluate and compare the performance of compressed sensing ensembles using four different sparse reconstruction algorithms, combined with two different classifier learning algorithms and two different training data manipulation techniques. We also give the comparison experiments of our method against another five state-of-the-art pruning methods. These experiments show that our method produces comparable or better accuracy, while being significantly faster than the compared methods. [Copyright &y& Elsevier]

Abstract: This paper presents variable-wise kernel hard clustering algorithms in the feature space in which dissimilarity measures are obtained as sums of squared distances between patterns and centroids computed individually for each variable by means of kernels. The methods proposed in this paper are supported by the fact that a kernel function can be written as a sum of kernel functions evaluated on each variable separately. The main advantage of this approach is that it allows the use of adaptive distances, which are suitable to learn the weights of the variables on each cluster, providing a better performance. Moreover, various partition and cluster interpretation tools are introduced. Experiments with synthetic and benchmark datasets show the usefulness of the proposed algorithms and the merit of the partition and cluster interpretation tools. [Copyright &y& Elsevier]

Abstract: The cosine similarity measure is often applied after discriminant analysis in pattern recognition. This paper first analyzes why the cosine similarity is preferred by establishing the connection between the cosine similarity based decision rule in the discriminant analysis framework and the Bayes decision rule for minimum error. The paper then investigates the challenges inherent of the cosine similarity and presents a new similarity that overcomes these challenges. The contributions of the paper are thus three-fold. First, the application of the cosine similarity after discriminant analysis is discovered to have its theoretical roots in the Bayes decision rule. Second, some inherent problems of the cosine similarity such as its inadequacy in addressing distance and angular measures are discussed. Finally, a new similarity measure, which overcomes the problems by integrating the absolute value of the angular measure and the l p norm (the distance measure), is presented to enhance pattern recognition performance. The effectiveness of the proposed new similarity measure in the discriminant analysis framework is evaluated using a large scale, grand challenge problem, namely, the Face Recognition Grand Challenge (FRGC) problem. Experimental results using 36,818 FRGC images on the most challenging FRGC experiment, the FRGC Experiment 4, show that the new similarity measure improves face recognition performance upon other popular similarity measures, such as the cosine similarity measure, the normalized correlation, and the Euclidean distance measure. [Copyright &y& Elsevier]

Abstract: Non-rigid 3D shape retrieval has become an active and important research topic in content-based 3D object retrieval. The aim of this paper is to measure and compare the performance of state-of-the-art methods for non-rigid 3D shape retrieval. The paper develops a new benchmark consisting of 600 non-rigid 3D watertight meshes, which are equally classified into 30 categories, to carry out experiments for 11 different algorithms, whose retrieval accuracies are evaluated using six commonly utilized measures. Models and evaluation tools of the new benchmark are publicly available on our web site . [Copyright &y& Elsevier]

Abstract: A novel method for the generation of synthetic on-line signatures based on the spectral analysis and the Kinematic Theory of rapid human movements, was presented in Part I of this series of two papers. In the present paper, the experimental framework used for the validation of the novel approach is described. The validation protocol, which uses different development and test sets in order to achieve unbiased results, includes the comparison of real and synthetic databases in terms of (i) visual appearance, (ii) statistical information, and (iii) performance evaluation of three competitive and totally different verification systems. The experimental results show the high similarity existing between synthetically generated and humanly produced samples, and the great potential of the method for the study of the signature trait. [Copyright &y& Elsevier]

Abstract: In this paper, we present the main results of the BioSecure Signature Evaluation Campaign (BSEC''2009). The objective of BSEC''2009 was to evaluate different online signature algorithms on two tasks: the first one aims at studying the influence of acquisition conditions (digitizing tablet or PDA) on systems'' performance; the second one aims at studying the impact of information content in signatures on systems'' performance. In BSEC''2009, the two BioSecure Data Sets DS2 and DS3 are used for tests, both containing data of the same 382 people, acquired respectively on a digitizing tablet and on a PDA. The results of the 12 systems involved in this evaluation campaign are reported and analyzed in detail in this paper. Experimental results reveal a 2.2% EER for skilled forgeries and a 0.51% EER for random forgeries on DS2; and a 4.97% EER for skilled forgeries and a 0.55% EER for random forgeries on DS3. [Copyright &y& Elsevier]

Highlights • Found and molded the relationship between the high-frequency signal of image carrier and steganalysis accuracy. • Posterior accuracy is proposed to evaluate the real accuracy of steganalysis and it can be used to improve steganalysis performance in the actual environment, because we can find a subset with a high accuracy(such as 100%) from the prediction images which have a relatively low accuracy (such as 80%). • Actually, the relationship between image carrier and steganalysis accuracy can even be used to improve steganography performance significantly. Abstract Steganalysis research is committed to distinguishing the steganography media from the normal one correctly. However, individual differences of carriers disturb the detection inevitably and greatly. Existing methods treat all detection results with the same confidence level, or prior accuracy, which may make the prior accuracy overestimate or underestimate the real result. This paper presents a novel performance evaluation method of steganalysis based on posterior accuracy. Adaptive Convolution Feature (ACF) is calculated by the adaptive convolution, then a quantitative value S based on the ACF is modeled to posterior testify and estimate the detection confidence of the image under test. By clustering of carrier noise, we classify the images which we believe they have a similar confidence of the same cluster. The distribution of ACF from each cluster indicates that the S value works well for confidence evaluation. The experimental results show that the S value can distinguish and identify high-confidence samples from the low-confidence, which will greatly improve the real performance of steganalysis. Besides, it improves the steganalysis accuracy of the whole image set by matching S values between training and prediction samples. [ABSTRACT FROM AUTHOR]

Abstract: This paper presents a novel face recognition method by means of fusing color, local spatial and global frequency information. Specifically, the proposed method fuses the multiple features derived from a hybrid color space, the Gabor image representation, the local binary patterns (LBP), and the discrete cosine transform (DCT) of the input image. The novelty of this paper is threefold. First, a hybrid color space, the RC r Q color space, is constructed by combining the R component image of the RGB color space and the chromatic component images, C r and Q, of the YC b C r and YIQ color spaces, respectively. The RC r Q hybrid color space, whose component images possess complementary characteristics, enhances the discriminating power for face recognition. Second, three effective image encoding methods are proposed for the component images in the RC r Q hybrid color space to extract features: (i) a patch-based Gabor image representation for the R component image, (ii) a multi-resolution LBP feature fusion scheme for the C r component image, and (iii) a component-based DCT multiple face encoding for the Q component image. Finally, at the decision level, the similarity matrices generated using the three component images in the RC r Q hybrid color space are fused using a weighted sum rule. Experiments on the Face Recognition Grand Challenge (FRGC) version 2 Experiment 4 show that the proposed method improves face recognition performance significantly. In particular, the proposed method achieves the face verification rate (ROC III curve) of 92.43%, at the false accept rate of 0.1%, compared to the FRGC baseline performance of 11.86% face verification rate at the same false accept rate. [Copyright &y& Elsevier]

Abstract: This paper presents a comparative study of three different strategies to improve the performance of particle filters, in the context of visual contour tracking: the unscented particle filter, the Rao-Blackwellized particle filter, and the partitioned sampling technique. The tracking problem analyzed is the joint estimation of the global and local transformation of the outline of a given target, represented following the active shape model approach. The main contributions of the paper are the novel adaptations of the considered techniques on this generic problem, and the quantitative assessment of their performance in extensive experimental work done. [Copyright &y& Elsevier]

Abstract: This paper investigates an information theoretic approach for formulating performance indices for the biometric authentication. Firstly, we formulate the constrained capacity, as a performance index for biometric authentication system for the finite number of users. Like Shannon capacity, constrained capacity is formulated using signal to noise ratio which is estimated from known statistics of users’ biometric information in the database. Constrained capacity of a user and of biometric system is fixed, given the database and the matching function. Experimental analysis using real palmprint and hand geometry images illustrates use of constrained capacity to estimate: (i) performance gains from the cohort information, (ii) the effective number of user-specific cohorts for a user and for the biometric system, (iii) information content of biometric features, and (iv) the performance of score level fusion rules for multimodal biometric system. Secondly, this paper investigates a rate-distortion framework for formulating false random correspondence probability as performance of a generic biometric. Our analysis concludes that constrained capacity can be a promising addition to performance of a biometric system. Similarly, individuality expressed as false random correspondence probability can be the performance index of a biometric trait. [Copyright &y& Elsevier]

Abstract: Reinforcement learning has been widely-used for applications in planning, control, and decision making. Rather than using instructive feedback as in supervised learning, reinforcement learning makes use of evaluative feedback to guide the learning process. In this paper, we formulate a pattern classification problem as a reinforcement learning problem. The problem is realized with a temporal difference method in a FALCON-R network. FALCON-R is constructed by integrating two basic FALCON-ART networks as function approximators, where one acts as a critic network (fuzzy predictor) and the other as an action network (fuzzy controller). This paper serves as a guideline in formulating a classification problem as a reinforcement learning problem using FALCON-R. The strengths of applying the reinforcement learning method to the pattern classification application are demonstrated. We show that such a system can converge faster, is able to escape from local minima, and has excellent disturbance rejection capability. [Copyright &y& Elsevier]

While standing as one of the most widely considered and successful supervised classification algorithms, the k -nearest Neighbor ( k NN) classifier generally depicts a poor efficiency due to being an instance-based method. In this sense, Approximated Similarity Search (ASS) stands as a possible alternative to improve those efficiency issues at the expense of typically lowering the performance of the classifier. In this paper we take as initial point an ASS strategy based on clustering. We then improve its performance by solving issues related to instances located close to the cluster boundaries by enlarging their size and considering the use of Deep Neural Networks for learning a suitable representation for the classification task at issue. Results using a collection of eight different datasets show that the combined use of these two strategies entails a significant improvement in the accuracy performance, with a considerable reduction in the number of distances needed to classify a sample in comparison to the basic k NN rule. [ABSTRACT FROM AUTHOR]

Multi-instance learning (MIL) is commonly used to classify a set of instances, also known as a bag, where labels for the training set are only available for each bag. Many MIL methods exist, but they often suffer from high computation complexity and the key information from MIL being ignored, which deteriorates the classification performance. Recently, locality-sensitive hashing (LSH), with its high scalability, has shown the ability in enhancing MIL performance. However, for these LSH-based methods, the fixed number of bits is used to represent each projected dimension, resulting in subtle information loss and the algorithm performance reduction. In this paper, we propose a self-adaptive LSH encoding method for MIL, termed as SALE. SALE uses LSH to generate the primary batches, followed by a self-adaptive process for reconstruction. Reconstructed bags are transformed into random super histograms (RSH) using an incomplete coding method, and then weighted through a scheme that takes advantage of key instances. These weighted RSHs are used to train the learning model. SALE efficiently deals with large MIL problems, due to its low complexity and RSH’s ability to exploit key information of MIL. Experiments demonstrate SALE’s good performance compared to state-of-the-art MIL methods. [ABSTRACT FROM AUTHOR]

Cross-ratio invariant is used widely in projective transformations for eye gaze estimation. Establishing a virtual plane projection is an important step to use this property. Most of traditional cross-ratio approaches only used fixed parameters to calculate the gazing point. This paper proposes gazing point dependent eye gazing estimation approach. Our contributions are three-folded. First, we model a dynamic virtual plane projection, which is tangent to the cornea of pupil, to estimate the position of the gazing point. Second, we introduce a two stage approach consisting of rough-to-precise framework for gazing point estimation based on the gazing point dependent virtual plane projection. Third, a heuristic strategy which contains off-line and on-line parameter learning for gazing point estimation is proposed. The experiment results show that our approach can significantly improve the gazing estimation performance with an average accuracy of 0.70°. [ABSTRACT FROM AUTHOR]

This paper presents two adaptive ensemble sampling approaches for imbalanced learning: one is the undersampling-based approach, and the other one is the oversampling-based approach, with the objectives of bias reduction and adaptive learning. Both of these two approaches are based on a novel class imbalance metric, termed generalized imbalance ratio (GIR), instead of the conventional sample size ratio. Specifically, these two sampling-based approaches adaptively split the imbalanced learning problem into multiple balanced learning subproblems in a probabilistic way, which forces the classifiers trained in the subproblems focus on those difficult to learn samples. In each subproblem, several weak classifiers are trained in a boosting manner. A final stronger classifier is further built by combining all these weak classifiers in a bagging manner. Extensive experiments are conducted on real-life UCI imbalanced data sets to evaluate the performance of the proposed methods. The superior performance demonstrates the effectiveness of the proposed methods and indicates wide potential applications in data mining. [ABSTRACT FROM AUTHOR]

This paper presents the novel hierarchical deep neural network (HDNN) for the general multivariate regression problem. The recent insight of deep neural network (DNN) is the deep architecture with large training data can bring the best performance in many research areas. The architecture design of our proposed HDNN focuses on both “depth” and “width” of artificial neural network. Specifically for the multivariate regression, HDNN consists of multiple subnets, which is empirically more powerful than DNN by using a divide and conquer strategy. The effectiveness of HDNN as the regression model is verified on two tasks, namely speech enhancement and Chinese handwriting recognition. For the speech enhancement task, our experiments show that HDNN significantly outperforms DNN in terms of perceptual evaluation of speech quality (PESQ), which is an objective measure highly correlated to subjective testing of listening quality. And for Chinese handwriting recognition task, as a nonlinear feature mapping function, we have a very interesting observation that DNN-based approach can not even bring performance gain while HDNN-based approach yields significant improvements of recognition accuracy. [ABSTRACT FROM AUTHOR]

This paper describes a method for robust offline writer identification. We propose to use RootSIFT descriptors computed densely at the script contours. GMM supervectors are used as encoding method to describe the characteristic handwriting of an individual scribe. GMM supervectors are created by adapting a background model to the distribution of local feature descriptors. Finally, we propose to use Exemplar-SVMs to train a document-specific similarity measure. We evaluate the method on three publicly available datasets (ICDAR / CVL / KHATT) and show that our method sets new performance standards on all three datasets. Additionally, we compare different feature sampling strategies as well as other encoding methods. [ABSTRACT FROM AUTHOR]

The number of methods available for classification of multi-label data has increased rapidly over recent years, yet relatively few links have been made with the related task of classification of sequential data. If labels indices are considered as time indices, the problems can often be seen as equivalent. In this paper we detect and elaborate on connections between multi-label methods and Markovian models, and study the suitability of multi-label methods for prediction in sequential data. From this study we draw upon the most suitable techniques from the area and develop two novel competitive approaches which can be applied to either kind of data. We carry out an empirical evaluation investigating performance on real-world sequential-prediction tasks: electricity demand, and route prediction. As well as showing that several popular multi-label algorithms are in fact easily applicable to sequencing tasks, our novel approaches, which benefit from a unified view of these areas, prove very competitive against established methods. [ABSTRACT FROM AUTHOR]

The dimensionality reduction methods based on linear embedding, such as neighborhood preserving embedding (NPE), sparsity preserving projections (SPP) and collaborative representation based projections (CRP), try to preserve a certain kind of linear representation for each sample after projection. However, in the transformed low-dimensional space, the linear relationship between the samples may be changed, which cannot make the linear representation-based classifiers, such as sparse representation-based classifier (SRC), to achieve higher recognition accuracy. In this paper, we propose a new linear dimensionality reduction algorithm, called Regularized Coplanar Discriminant Analysis (RCDA) to address this problem. It simultaneously seeks a linear projection matrix and some linear representation coefficients that make the samples from the same class coplanar and the samples from different classes not coplanar. The proposed regularization term balances the bias from the optimal linear representation and that from the class mean to avoid overfitting the training data, and overcomes the matrix singularity in solving the linear representation coefficients. An alternative optimization approach is proposed to solve the RCDA model. Experiments are done on several benchmark face databases and hyperspectral image databases, and results show that RCDA can obtain better performance than other dimensionality reduction methods. [ABSTRACT FROM AUTHOR]

Occlusion poses as a critical challenge in computer vision for a long time. Recently, the technique of synthetic aperture photography using a camera array has been regarded as a promising way to address the problem of occluded object imaging. Unfortunately, the expensive cost of a standard camera array system with the required calibration procedure still limits the widespread popularity of this technique. To make synthetic aperture photography practical, in this paper, we propose a novel approach based on a mobile camera-IMU system moving on a planar surface. The contributions of our work include: (1) with geometric property assumptions, we mathematically prove that the transformation matrices for synthetic aperture image generation can be factorized into an infinite homography and a shift; (2) the IMU data are utilized to estimate the homography by selecting the plane at infinity as the reference plane, which is scene independent and does not require any calibration; (3) compared to the conventional camera array based approaches, the proposed method effectively improves the solution for synthetic aperture images capturing and generation by using a more flexible, convenient and inexpensive system. Extensive experimental results with qualitative and quantitative evaluations have demonstrated the see-through-occlusion performance with satisfactory accuracy of the proposed approach in different scenarios. [ABSTRACT FROM AUTHOR]

Biclustering represents an intrinsically complex problem, where the aim is to perform a simultaneous row- and column-clustering of a given data matrix. Some recent approaches model this problem using factor graphs, so to exploit their ability to open the door to efficient optimization approaches for well designed function decompositions. However, while such models provide promising results, they do not scale to data matrices of reasonable size. In this paper, we take a step towards addressing this issue, by proposing a novel approach to biclustering based on factor graphs, which yields high quality solutions and scales more favorably than previous methods. Specifically, we cast biclustering as the sequential search for a single bicluster, and propose a binary and compact factor graph that can be solved efficiently using the max-sum algorithm. The proposed approach has been tested and compared with state-of-the-art methods on four datasets (two synthetic and two real world data), providing encouraging results with respect both to previous approaches based on factor graphs and to other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Innovative technologies, such as 3D depth cameras, promote the development of natural interaction applications in many domains among large audiences. In this context, supervised machine learning techniques have been proved to be a flexible and robust approach to perform high level gesture recognition from 3D joints provided by these depth cameras. This paper proposes a lightweight approach to recognize gestures with Kinect by utilizing approximate string matching. The proposed approach encodes the movements of the joints as sequences of characters in order to simplify the gesture recognition as a widely studied string matching problem. We evaluated our approach by applying other widespread used techniques in the research field. The experimental evaluations show that the proposed approach can obtain relatively high performance in comparison with the state-of-the-art machine learning techniques. These findings provide further evidence that our approach could be a viable strategy for recognizing gestures, even in devices with medium and low processing capability (e.g., smartphones, tablets, etc.). [ABSTRACT FROM AUTHOR]

Despite emerging as a prominent choice to serve the security concerns of person authentication applications, unimodal biometric systems are vulnerable to spoof attacks. Multimodal biometric systems can effectively minimize spoof attacks while improving the overall performance. In this paper, we present a multimodal system based on two modalities derived from multi instance fingerprint acquisition viz. fingerprint and the associated time dynamics. Extensive user verification and spoof resistance experiments conducted on virtual multimodal databases, created by combining ATVS and LivDet-13 fingerprint databases each with fingerprint dynamics database. Fusion is performed at match score level using sum and weighted sum rules. The empirical results demonstrate spoof resistance of the proposed multimodal system with significant performance improvement over unimodal and multi-instance fingerprint recognition systems. The performance of the proposed system is evaluated on well-known metrics like Detection Error Trade-off (DET) curves, equal error rate (EER), and Area Under the Curve (AUC). [ABSTRACT FROM AUTHOR]

A competition in unsupervised color image segmentation took place in conjunction with the 22nd International Conference on Pattern Recognition (ICPR 2014). It aimed to promote evaluation of unsupervised color image segmentation algorithms using publicly available data sets, and to allow for any subsequent methods to be easily evaluated and compared with the results of the contested methods under identical conditions. Our comparison of different methods is based on the standard methodology of performance assessment using an on-line verification server. We present in this paper the evaluation of the top six results submitted to the ICPR 2014 contest in unsupervised color image segmentation and compare them with 11 other state-of-the-art unsupervised image segmenters. [ABSTRACT FROM AUTHOR]

Disparity refinement is the final step but also the timing bottleneck of stereo matching due to its high computational complexity. Weighted media filter refinement method and non-local refinement method are two typical refinement methods with O ( N ) computational complexity for each pixel where N indicates the maximum disparity. This paper presents an O (1) disparity refinement method based on belief aggregation and belief propagation . The aggregated belief , which means the possibility of correct disparity value, is efficiently computed on a minimum spanning tree first, and then the belief aggregation is fast performed on another minimum spanning tree in two sequential passes (first from leaf nodes to root, then from root to leaf nodes). Only 2 additions and 4 multiplications are required for each pixel at all disparity levels, so the computational complexity is O (1). Performance evaluation on Middlebury data sets shows that the proposed method has good performances both in accuracy and speed. [ABSTRACT FROM AUTHOR]

In this paper, we present an Image-to-Class Dynamic Time Warping (I2C-DTW) approach for the recognition of both 3D static hand gestures and 3D hand trajectory gestures. Our contribution is twofold. First, we propose a technique to compute the image-to-class dynamic time warping distance instead of the Image-to-Image distance. By doing so, we obtain better generalization capability using the Image-to-Class distance than the Image-to-Image distance. Second, we propose a compositional model called fingerlets for static gesture representation, and a compositional model called strokelets for trajectory gesture representation. The compositional models make it possible to compute the DTW distance between a data sample and a gesture category. We have evaluated the static gesture recognition performance on several public 3D hand gesture datasets. For better evaluating the performance on trajectory gesture recognition, we collected a 3D hand trajectory gesture dataset, called UESTC-HTG, using a Kinect device. The experiment results show that the proposed I2C-DTW approach significantly improves the recognition accuracy on both static gestures and trajectory gestures. [ABSTRACT FROM AUTHOR]

Recently in Dufrenois [1] , a new Fisher type contrast measure has been proposed to extract a target population in a dataset contaminated by outliers. Although mathematically sound, this work presents some further shortcomings in both the formalism and the field of use. First, we propose to re-express this problem from the formalism of proximal support vector machines as introduced in Mangasarian and Wild [2] . This change is far from harmless since it introduces a suited writing for solving the problem. Another limiting factor of the method is that its performance relies on the assumption that the density between the target and outliers are different. This consideration can easily prove to be over-optimistic for real world datasets making the method unreliable, at least directly. The computation of the decision boundary is a time consuming part of the algorithm since it is based on solving a generalized eigenvalue problem (GEP). This method is therefore limited to medium sized data sets. In this paper, we propose appropriate strategies to unlock all these shortcomings and fully benefit from the interest of the approach. Firstly, we show under some conditions that generating appropriate artificial outliers allows to stay within the constraints of the method and thus enlarges the conditions of use. Secondly, we show that the GEP can be advantageously replaced by a conjugate gradient solution (CG) significantly decreasing the computational cost. Lastly, the proposed algorithm is compared with recent novelty detectors on synthetic and real datasets. [ABSTRACT FROM AUTHOR]

This paper presents kernel-based hard clustering methods with kernelization of the metric and automatic weighting of the variables. The proposed methodology is supported by the fact that a kernel function can be written as a sum of kernels evaluated separately on each variable. Thus, in the proposed algorithms dissimilarity measures are obtained as sums of Euclidean distances between patterns and centroids computed individually for each variable by means of kernels. The main advantage of this approach over the conventional approach is that it allows the use of kernelized adaptive distances, which are suitable to learn the weights of the variables dynamically, improving the performance of the algorithms. Moreover, various partition and cluster interpretation tools are introduced. Experiments with a number of benchmark datasets corroborate the usefulness of the proposed algorithms and the merit of the partition and cluster interpretation tools. [ABSTRACT FROM AUTHOR]

Categorical data clustering is an important subject in pattern recognition. Currently, subspace clustering of categorical data remains an open problem due to the difficulties in estimating attribute interestingness according to the statistics of categories in clusters. In this paper, a new algorithm is proposed for clustering categorical data with a novel soft feature-selection scheme, by which each categorical attribute is automatically assigned a weight that correlates with the smoothed dispersion of the categories in a cluster. In the proposed algorithm, dissimilarity between categorical data objects is measured using a probabilistic distance function, based on kernel density estimation for categorical attributes. We also make use of the probabilistic distances to define a cluster validity index for estimating the number of categorical clusters. The suitability of the proposal is demonstrated in an empirical study done with some widely used real-world data sets and synthetic data sets, and the results show its outstanding performance. [ABSTRACT FROM AUTHOR]

Mislabeled examples in the training data can severely affect the performance of supervised classifiers. In this paper, we present an approach to remove any mislabeled examples in the dataset by selecting suspicious examples as targets for inspection. We show that the large margin and soft margin principles used in support vector machines (SVM) have the characteristic of capturing the mislabeled examples as support vectors. Experimental results on two character recognition datasets show that one-class and two-class SVMs are able to capture around 85% and 99% of label noise examples, respectively, as their support vectors. We propose another new method that iteratively builds two-class SVM classifiers on the non-support vector examples from the training data followed by an expert manually verifying the support vectors based on their classification score to identify any mislabeled examples. We show that this method reduces the number of examples to be reviewed, as well as providing parameter independence of this method, through experimental results on four data sets. So, by (re-)examining the labels of the selective support vectors, most noise can be removed. This can be quite advantageous when rapidly building a labeled data set. [ABSTRACT FROM AUTHOR]

The problem of human detection is to automatically locate people in an image or video sequence and has been actively researched in the past decade. This paper aims to provide a comprehensive survey on the recent development and challenges of human detection. Different from previous surveys, this survey is organised in the thread of human object descriptors. This approach has advantages in providing a thorough analysis of the state-of-the-art human detection methods and a guide to the selection of appropriate methods in practical applications. In addition, challenges such as occlusion and real-time human detection are analysed. The commonly used evaluation of human detection methods such as the datasets, tools, and performance measures are presented and future research directions are highlighted. [ABSTRACT FROM AUTHOR]

In the area of pattern recognition, Low Rank Representation (LRR) is an efficient method in recovering the subspace structure of the dataset. However, LRR is unsupervised. Without any label information, LRR constructs an informative graph which is then combined with the mature graph-based semi-supervised learning (GSSL) framework to complete the classification task. In this paper, we propose a new low rank learning method which constructs the low rank representation matrix utilizing label information to obtain a more informative graph. This method integrates the low rank graph construction and the label information propagation processes together. Thus the optimization of the low rank representation and the soft label prediction function are calculated iteratively at the same time. We name this method as Semi-Supervised Low Rank Learning (SSLRL). It enhanced the classification performance of traditional LRR-Graph based SSL by 5–30% and the running time is reduced from hundreds to less than ten seconds. Based on this method, a new outlier detection strategy is presented. This strategy succeeds with an AUC of at least 93% even if the detection condition of LRR is not satisfied. The effectiveness of SSLRL is demonstrated in semi-supervised classification, outlier detection, and salient detection tasks. These extensive experimental results highlight the outperforming of our method over state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Local image features are routinely used in state-of-the-art methods to solve many computer vision problems like image retrieval, classification, or 3D registration. As the applications become more complex, the research for better visual features is still active. In this paper we present a feature detector that exploits the inherent geometry of sampled image edges using α - shapes . We propose a novel edge sampling scheme that exploits local shape and investigate different triangulations of sampled points. We also introduce a novel approach to represent the anisotropy in a triangulation along with different feature selection methods. Our detector provides a small number of distinctive features that is ideal for large scale applications, while achieving competitive performance in a series of matching and retrieval experiments. [ABSTRACT FROM AUTHOR]

Recently, a multi-view dictionary learning (DL) technique has received much attention. Although some multi-view DL methods have been presented, they suffer from the problem of performance degeneration when large noise exists in multiple views. In this paper, we propose a novel multi-view DL approach named multi-view low-rank DL (MLDL) for image classification. Specifically, inspired by the low-rank matrix recovery theory, we provide a multi-view dictionary low-rank regularization term to solve the noise problem. We further design a structural incoherence constraint for multi-view DL, such that redundancy among dictionaries of different views can be reduced. In addition, to enhance efficiency of the classification procedure, we design a classification scheme for MLDL, which is based on the idea of collaborative representation based classification. We apply MLDL for face recognition, object classification and digit classification tasks. Experimental results demonstrate the effectiveness and efficiency of the proposed approach. [ABSTRACT FROM AUTHOR]

This paper presents a robust text detection approach based on color-enhanced contrasting extremal region (CER) and neural networks. Given a color natural scene image, six component-trees are built from its grayscale image, hue and saturation channel images in a perception-based illumination invariant color space, and their inverted images, respectively. From each component-tree, color-enhanced CERs are extracted as character candidates. By using a “divide-and-conquer” strategy, each candidate image patch is labeled reliably by rules as one of five types, namely, Long, Thin, Fill, Square-large and Square-small, and classified as text or non-text by a corresponding neural network, which is trained by an ambiguity-free learning strategy. After pruning unambiguous non-text components, repeating components in each component-tree are pruned further. Remaining components are then grouped into candidate text-lines and verified by another set of neural networks. Finally, results from six component-trees are combined, and a post-processing step is used to recover lost characters. Our proposed method achieves superior performance on both ICDAR-2011 and ICDAR-2013 “Reading Text in Scene Images” test sets. [ABSTRACT FROM AUTHOR]

Classification is an essential task for predicting the class values of new instances. Both k -fold and leave-one-out cross validation are very popular for evaluating the performance of classification algorithms. Many data mining literatures introduce the operations for these two kinds of cross validation and the statistical methods that can be used to analyze the resulting accuracies of algorithms, while those contents are generally not all consistent. Analysts can therefore be confused in performing a cross validation procedure. In this paper, the independence assumptions in cross validation are introduced, and the circumstances that satisfy the assumptions are also addressed. The independence assumptions are then used to derive the sampling distributions of the point estimators for k -fold and leave-one-out cross validation. The cross validation procedure to have such sampling distributions is discussed to provide new insights in evaluating the performance of classification algorithms. [ABSTRACT FROM AUTHOR]

In this paper we present an efficient segmentation-free word spotting method, applied in the context of historical document collections, that follows the query-by-example paradigm. We use a patch-based framework where local patches are described by a bag-of-visual-words model powered by SIFT descriptors. By projecting the patch descriptors to a topic space with the latent semantic analysis technique and compressing the descriptors with the product quantization method, we are able to efficiently index the document information both in terms of memory and time. The proposed method is evaluated using four different collections of historical documents achieving good performances on both handwritten and typewritten scenarios. The yielded performances outperform the recent state-of-the-art keyword spotting approaches. [ABSTRACT FROM AUTHOR]

The multi-target tracking problem is challenging when there exist occlusions, tracking failures of the detector and severe interferences between detections. In this paper, we propose a novel detection based tracking method that links detections into tracklets and further forms long trajectories. Unlike many previous hierarchical frameworks which split the data association into two separate optimization problems (linking detections locally and linking tracklets globally), we introduce a unified algorithm that can automatically relearn the trajectory models from the local and global information for finding the joint optimal assignment. In each temporal window, the trajectory models are initialized by the local information to link those easy-to-connect detections into a set of tracklets. Then the trajectory models are updated by the reliable tracklets and reused to link separated tracklets into long trajectories. We iteratively update the trajectory models by more information from more frames until the result converges. The iterative process gradually improves the accuracy of the trajectory models, which in turn improves the target ID inferences for all detections by the MRF model. Experiment results revealed that our proposed method achieved state-of-the-art multi-target tracking performance. [ABSTRACT FROM AUTHOR]

Real-world applications demand effective methods to estimate the class distribution of a sample. In many domains, this is more productive than seeking individual predictions. At a first glance, the straightforward conclusion could be that this task, recently identified as quantification, is as simple as counting the predictions of a classifier. However, due to natural distribution changes occurring in real-world problems, this solution is unsatisfactory. Moreover, current quantification models based on classifiers present the drawback of being trained with loss functions aimed at classification rather than quantification. Other recent attempts to address this issue suffer certain limitations regarding reliability, measured in terms of classification abilities. This paper presents a learning method that optimizes an alternative metric that combines simultaneously quantification and classification performance. Our proposal offers a new framework that allows the construction of binary quantifiers that are able to accurately estimate the proportion of positives, based on models with reliable classification abilities. [ABSTRACT FROM AUTHOR]

Receiver operating characteristic (ROC) analysis is a standard methodology to evaluate the performance of a binary classification system. The area under the ROC curve (AUC) is a performance metric that summarizes how well a classifier separates two classes. Traditional AUC optimization techniques are supervised learning methods that utilize only labeled data (i.e., the true class is known for all data) to train the classifiers. In this work, inspired by semi-supervised and transductive learning, we propose two new AUC optimization algorithms hereby referred to as semi-supervised learning receiver operating characteristic (SSLROC) algorithms, which utilize unlabeled test samples in classifier training to maximize AUC. Unlabeled samples are incorporated into the AUC optimization process, and their ranking relationships to labeled positive and negative training samples are considered as optimization constraints. The introduced test samples will cause the learned decision boundary in a multi-dimensional feature space to adapt not only to the distribution of labeled training data, but also to the distribution of unlabeled test data. We formulate the semi-supervised AUC optimization problem as a semi-definite programming problem based on the margin maximization theory. The proposed methods SSLROC1 (1-norm) and SSLROC2 (2-norm) were evaluated using 34 (determined by power analysis) randomly selected datasets from the University of California, Irvine machine learning repository. Wilcoxon signed rank tests showed that the proposed methods achieved significant improvement compared with state-of-the-art methods. The proposed methods were also applied to a CT colonography dataset for colonic polyp classification and showed promising results. 1 1 Matlab code of the proposed methods will be released on http://clinicalcenter.nih.gov/drd/summers.html once the paper is published. [ABSTRACT FROM AUTHOR]

This paper presents a visual object tracking system which is tolerant to external imaging factors such as illumination, scale, rotation, occlusion and background changes. Specifically, an integration of an online version of total-error-rate minimization based projection network with an observation model of particle filter is proposed to effectively distinguish between the target object and the background. A re-weighting technique is proposed to stabilize the sampling of particle filter for stochastic propagation. For self-adaptation, an automatic updating scheme and extraction of training samples are proposed to adjust to system changes online. Our qualitative and quantitative experiments on 16 public video sequences show convincing performances in terms of tracking accuracy and computational efficiency over competing state-of-the-art algorithms. [ABSTRACT FROM AUTHOR]

Abstract: Word recognition systems use a lexicon to guide the recognition process in order to improve the recognition rate. However, as the lexicon grows, the computation time increases. In this paper, we present the Arabic word descriptor (AWD) for Arabic word shape indexing and lexicon reduction in handwritten documents. It is formed in two stages. First, the structural descriptor (SD) is computed for each connected component (CC) of the word image. It describes the CC shape using the bag-of-words model, where each visual word represents a different local shape structure, extracted from the image with filters of different patterns and scales. Then, the AWD is formed by sorting and normalizing the SDs. This emphasizes the symbolic features of Arabic words, such as subwords and diacritics, without performing layout segmentation. In the context of lexicon reduction, the AWD is used to index a reference database. Given a query image, the reduced lexicon is obtained from the labels of the first entries in the indexed database. This framework has been tested on Arabic word databases. It has a low computational overhead, while providing a compact descriptor, with state-of-the-art results for lexicon reduction on the Ibn Sina and IFN/ENIT databases. [Copyright &y& Elsevier]

Abstract: This paper first proposes a new type of single-output Chebyshev-polynomial feed-forward neural network (SOCPNN) for pattern classification. A new type of multi-output Chebyshev-polynomial feed-forward neural network (MOCPNN) is then proposed based on such an SOCPNN. Compared with multi-layer perceptron, the proposed SOCPNN and MOCPNN have lower computational complexity and superior performance, substantiated by both theoretical analyses and numerical verifications. In addition, two weight-and-structure-determination (WASD) algorithms, one for the SOCPNN and another for the MOCPNN, are proposed for pattern classification. These WASD algorithms can determine the weights and structures of the proposed neural networks efficiently and automatically. Comparative experimental results based on different real-world classification datasets with and without added noise prove that the proposed SOCPNN and MOCPNN have high accuracy, and that the MOCPNN has strong robustness in pattern classification when equipped with WASD algorithms. [Copyright &y& Elsevier]

Abstract: Random Forests receive much attention from researchers because of their excellent performance. As Breiman suggested, the performance of Random Forests depends on the strength of the weak learners in the forests and the diversity among them. However, in the literature, many researchers only considered pre-processing of the data or post-processing of the Random Forests models. In this paper, we propose a new method to increase the diversity of each tree in the forests and thereby improve the overall accuracy. During the training process of each individual tree in the forest, different rotation spaces are concatenated into a higher space at the root node. Then the best split is exhaustively searched within this higher space. The location where the best split lies decides which rotation method to be used for all subsequent nodes. The performance of the proposed method here is evaluated on 42 benchmark data sets from various research fields and compared with the standard Random Forests. The results show that the proposed method improves the performance of the Random Forests in most cases. [Copyright &y& Elsevier]

Abstract: In this paper, we proposed multi-factors correlation (MFC) to describe the image, structure element correlation (SEC), gradient value correlation (GVC) and gradient direction correlation (GDC). At first, the RGB color space image is converted to a bitmap image and a mean color component image utilizing the block truncation coding (BTC). Then, three correlations will be used to extract the image feature. The structure elements can effectively represent the bitmap which is generated by BTC, and SEC can effectively denote the bitmap׳s structure and the correlation of the block in the bitmap. GVC and GDC can effectively denote the gradient relation, which is computed by a mean color component image. Formed by SEC, GVC and GDC, the image feature vectors can effectively represent the image. In the end, the results demonstrate that the method has better performance than other image retrieval methods in the experiment. [Copyright &y& Elsevier]

Abstract: Human Epithelial type 2 (HEp-2) cells play an important role in the diagnosis of autoimmune disorder. Traditional approach relies on specialists to observe HEp-2 slides via the fluorescence microscope, which suffers from a number of shortcomings like being subjective and labor intensive. Pattern recognition techniques have been recently introduced to this research issue to make the process automatic. However, performances of current systems available in literature are not satisfying. We propose in this paper a framework using intensity order pooling based gradient feature and bag of words for HEp-2 classification. By pooling the gradient features based on the intensity orders of local grid points, the pooled feature is rotationally invariant without requirement of orientation estimation. The proposed approach was fully tested using publicly available ICPR dataset and our own SZU dataset. Experimental results show that the propose method significantly outperformed widely used SIFT feature and the winner of ICPR contest 2012. Encouraging 100% image level accuracy was achieved on the SZU dataset. [Copyright &y& Elsevier]

Abstract: Autoimmune diseases are proven to be connected with the occurrence of autoantibodies in patient serum. Antinuclear autoantibodies (ANAs) identification can be accomplished in a laboratory using indirect immunofluorescence (IIF) imaging. In this paper a system for automatic classification of staining patterns on HEp-2 fluorescence images is proposed. Our method utilizes two descriptors in order to encode gradient and textural characteristics of the depicted patterns. Along with distribution of SIFT features, we propose the new GoC-LBP descriptor based on co-occurrences of uniform Local Binary Patterns along directions dictated by the orientation of local gradient. At a second stage, the descriptors are fused while creating a dissimilarity representation of an image. A powerful classification scheme is incorporated, utilizing a discriminative sparse representation-based scheme for the classification. Experiments were conducted using a publicly available dataset, comparing the obtained performance to recently reported results of a relevant contest, demonstrating the effectives of the proposed method. [Copyright &y& Elsevier]