Search

Highlights • A new patch-based method is proposed for single sample per person face recognition. • A Fisher-like criterion is designed to extract discriminant information of patches. • Two distance metrics are introduced simultaneously to enhance robustness. • A fusion strategy called joint majority voting is developed for recognition. Abstract Single sample per person face recognition is one of the most challenging problems in face recognition (FR), where only single sample per person (SSPP) is enrolled in the gallery set for training. Although the existing patch-based methods have achieved great success in FR with SSPP, they still have limitations in feature extraction and identification stages when handling complex facial variations. In this work, we propose a new patch-based method called Robust Heterogeneous Discriminative Analysis (RHDA), for FR with SSPP. To enhance the robustness against complex facial variations, we first present a new graph-based Fisher-like criterion, which incorporates two manifold embeddings, to learn heterogeneous discriminative representations of image patches. Specifically, for each patch, the Fisher-like criterion is able to preserve the reconstruction relationship of neighboring patches from the same person, while suppressing the similarities between neighboring patches from the different persons. Then, we introduce two distance metrics, i.e., patch-to-patch distance and patch-to-manifold distance, and develop a fusion strategy to combine the recognition outputs of above two distance metrics via a joint majority voting for identification. Experimental results on various benchmark datasets demonstrate the effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Highlights • A very high spatial resolution images land-use classification scheme is proposed. • The proposed method relies on dictionaries of deep features. • These dictionaries are very discriminative and compact. • Likelihoods are linked to the sparse representation approach. • The proposed method can be competitive in a comparison with the state-of-the-art. Abstract Land-use classification in very high spatial resolution images is critical in the remote sensing field. Consequently, remarkable efforts have been conducted towards developing increasingly accurate approaches for this task. In recent years, deep learning has emerged as a dominant paradigm for machine learning, and methodologies based on deep convolutional neural networks have received particular attention from the remote sensing community. These methods typically utilize transfer learning and/or data augmentation to accommodate a small number of labeled images in the publicly available datasets in this field. However, they typically require powerful computers and/or a long time for training. In this work, we propose a simple and novel method for land-use classification in very high spatial resolution images, which efficiently combines transfer learning with a sparse representation. Specifically, the proposed method performs the classification of land-use scenes using a modified version of the well-known sparse representation-based classification method. While this method directly uses the training images to form dictionaries, which are employed to classify test images, our method utilizes a pre-trained deep convolutional neural network and the Gaussian mixture model to generate more robust and compact "dictionaries of deep features." The effectiveness of the proposed method was evaluated on two publicly available datasets: UC Merced and Brazilian Cerrado–Savana. The experimental results suggest that our method can potentially outperform state-of-the-art techniques for land-use classification in very high spatial resolution images. [ABSTRACT FROM AUTHOR]

Highlights • The novel global weighted SPF (GWSPF) is defined by introducing the normalized global minimum absolute differences as the coefficients of global inner and outer region fitting centers. • The novel local weighted SPF (LWSPF) is similarly defined by introducing the normalized local minimum absolute differences as the coefficients of local inner and outer region fitting centers. • The global and local within-class variances of the image are used to weight the GWSPF and the LWSPF, which can automatically adjust the effect degrees of the GWSPF and the LWSPF. • The proposed model is superior to state-of-the-art active contour models in segmentation accuracy, in addition, it is robust to the initial curve. Abstract This paper proposes a new global and local weighted signed pressure force (SPF) based active contour model (ACM) to segment various types of images. First, by introducing the normalized global minimum absolute differences as the coefficients of global inner and outer region fitting centers, a new global weighted SPF (GWSPF) is defined, which makes the best of the difference information of inner and outer regions and improves segmentation performance. Second, by introducing the normalized local minimum absolute differences as the coefficients of local inner and outer region fitting centers similarly, a new local weighted SPF (LWSPF) is defined and added to the above global weighted SPF. Third, the global and local within-class variances of the image are used to weight the GWSPF and the LWSPF, which can automatically adjust the effect degrees of the GWSPF and the LWSPF. Experiments on many kinds of real-world images have validated that the proposed model is superior to popular ACMs in segmentation accuracy, in addition, it is robust to the initial curve. [ABSTRACT FROM AUTHOR]

Highlights • A new data-dependent method for determining the trimming level in robust clustering. • We monitor the cluster structure as the trimming level changes. • The robust clustering constrains the ratio of eigenvalues of dispersion matrices. • The "car-bike" plot exhibits cluster stability as the eigenvalue constraint varies. • Sizable bibliography relating the work to that in the machine learning and pattern recognition fields. Abstract Monitoring the properties of single sample robust analyses of multivariate data as a function of breakdown point or efficiency leads to the adaptive choice of the best values of these parameters, eliminating arbitrary decisions about their values and so increasing the quality of estimators. Monitoring the trimming proportion in robust cluster analysis likewise leads to improved estimators. We illustrate these procedures on a sample of 424 cows with bovine phlegmon. For clustering we use a method which includes constraints on the eigenvalues of the dispersion matrices, so avoiding thread shaped clusters. The "car-bike" plot reveals the stability of clustering as the trimming level changes. The pattern of clusters and outliers alters appreciably for low levels of trimming. [ABSTRACT FROM AUTHOR]

Highlights • We present a method for writer identification based on single handwritten word images. • High-level features learned from the auxiliary task are transferred to the task of writer identification in a multi-task learning framework. • We evaluate the deep adaptive learning with three auxiliary tasks: word content, word length and character attribute recognition. Abstract There are two types of information in each handwritten word image: explicit information which can be easily read or derived directly, such as lexical content or word length, and implicit attributes such as the author's identity. Whether features learned by a neural network for one task can be used for another task remains an open question. In this paper, we present a deep adaptive learning method for writer identification based on single-word images using multi-task learning. An auxiliary task is added to the training process to enforce the emergence of reusable features. Our proposed method transfers the benefits of the learned features of a convolutional neural network from an auxiliary task such as explicit content recognition to the main task of writer identification in a single procedure. Specifically, we propose a new adaptive convolutional layer to exploit the learned deep features. A multi-task neural network with one or several adaptive convolutional layers is trained end-to-end, to exploit robust generic features for a specific main task, i.e., writer identification. Three auxiliary tasks, corresponding to three explicit attributes of handwritten word images (lexical content, word length and character attributes), are evaluated. Experimental results on two benchmark datasets show that the proposed deep adaptive learning method can improve the performance of writer identification based on single-word images, compared to non-adaptive and simple linear-adaptive approaches. [ABSTRACT FROM AUTHOR]

Highlights • We propose a multi-view discriminant learning based sparse representation method to explore group similarity in the multi-feature space. • The proposed method makes use of unreliable observation group to achieve multi-view fusion and makes different observation groups more group discriminative. • The proposed sparse representation method is incorporated into a particle filter based framework to achieve robust visual tracking. • Our method can achieve a better tracking performance than state-of-the-art tracking methods do. Abstract The multi-view sparse representation based visual tracking has attracted increasing attention because the sparse representations of different object features can complement with each other. Since the robustness of different object features is actually not the same in challenging video sequences, it may contain unreliable features (the features with low robustness) in multi-view sparse representation. In this case, how to highlight the useful information of unreliable features for proper multi-feature fusion has become a tough work. To solve this problem, we propose a multi-view discriminant sparse representation method for robust visual tracking, in which we firstly divide the multi-view observations into different groups, and then estimate the sparse representations of multi-view group projections for calculating the observation likelihood. The advantages of the proposed sparse representation method are two-folds: 1) It can properly fuse the observation groups with reliable and unreliable features by using an online updated discriminant matrix to explore the group similarity in multi-feature space. 2) It introduces a nonlocal regularizer to enforce the spatial smoothness among the sparse representations of different group projections, which can enhance the robustness of multi-view sparse representation. Experimental results show that our method can achieve a better tracking performance than state-of-the-art tracking methods do. [ABSTRACT FROM AUTHOR]

Highlights • We propose the bowler-hat transform - a new multiscale vessel enhancement approach based on mathematical morphology. • The method combines different structuring elements to detect innate features of vessel-like structures. • We evaluate the proposed method qualitatively and quantitatively, and compare it with the existing, state-of-the-art methods using both synthetic and real datasets. • The proposed method achieves high-quality vessel-like structure enhancement in both synthetic examples and in clinically relevant retinal images, and is shown to be able to detect fine vessels while remaining robust at junctions. Abstract Enhancement, followed by segmentation, quantification and modelling of blood vessels in retinal images plays an essential role in computer-aided retinopathy diagnosis. In this paper, we introduce the bowler-hat transform method a new approach based on mathematical morphology for vessel enhancement. The proposed method combines different structuring elements to detect innate features of vessel-like structures. We evaluate the proposed method qualitatively and quantitatively and compare it with the state-of-the-art methods using both synthetic and real datasets. Our results establish that the proposed method achieves high-quality vessel-like structure enhancement in both synthetic examples and clinically relevant retinal images. The bowler-hat transform is shown to be able to detect fine vessels while still remaining robust at junctions. [ABSTRACT FROM AUTHOR]

Highlights • A Bayesian denoising hashing method is proposed for robust image retrieval. • We impose extra constraints in the variational Bayes framework to learn robust hash code. • Our method can be combined with many traditional hashing algorithms to achieve state-of-the-art performance on real-world datasets. Abstract Learning to hash is one of the most popular techniques in image retrieval, but few work investigates its robustness to noise corrupted images in which the unknown pattern of noise would heavily deteriorate the performance. To deal with this issue, we present in this paper a Bayesian denoising hashing algorithm whose output can be regarded a denoised version of the input hash code. We show that our method essentially seeks to reconstruct a new but more robust hash code by preserving the original input information while imposing extra constraints so as to correct the corrupted bits. We optimized this model in variational Bayes framework which has a closed-form update in each iteration that is more efficient than numerical optimization. Furthermore, our method can be added at the top of any original hashing layer, serving as a post-processing denoising layer with no change to previous training procedure. Experiments on three popular datasets demonstrate that the proposed method yields robust and meaningful hash code, which significantly improves the performance of state-of-the-art hash learning methods on challenging tasks such as large-scale natural image retrieval and retrieval with corrupted images. [ABSTRACT FROM AUTHOR]

Highlights • A robust landmark-based algorithm for finding the geometric median (group average) of a group of shapes with heavy outliers is proposed. • The underlying Log and Exp maps of the proposed algorithm belong to a class of efficient geodesic shooting algorithms for template matching. • Once the group average is found, the Hamiltonian metric computed between the group average and each member of the group can be used for clustering analysis. • The proposed algorithms, as a tool of feature extraction, extract information about momentum at each landmark during the process of finding the geometric median. • The local momenta found by the proposed algorithm are useful for some classification purposes that cannot be achieved by using the landmark locations. Abstract We introduce a class of mathematical algorithms with the aim of establishing a framework of finding a group average and extracting prominent features in a group of landmark represented shapes or image templates. A group average is an estimator that is said to best represent the common features of the group being studied. The proposed algorithms, as a tool of feature extraction, extract information about momentum at each landmark through the process of template matching. Once the convergence criterion is satisfied numerically, the algorithms produce a group average and a local coordinate system for each member of the observing group, in terms of the residual momentum. We present several examples to illustrate the use of the proposed algorithms for finding a group average. Using the metrics computed between the group average and each member of the group, we successfully run a cluster analysis for datasets that contain a heavy percentage of outliers. Finally, we apply the collected residual momenta computed in the proposed algorithms in some statistical methods to demonstrate a potential application of the algorithms for detecting structure abnormality. [ABSTRACT FROM AUTHOR]

Highlights • Six convnet fusion architectures and their adaptations are investigated. • IAF R-CNN model for multispectral pedestrian detection is proposed. • An illumination-aware weighting mechanism is introduced. • A new state-of-the-art result on KAIST Benchmark is reported. Abstract Multispectral images of color-thermal pairs have shown more effective than a single color channel for pedestrian detection, especially under challenging illumination conditions. However, there is still a lack of studies on how to fuse the two modalities effectively. In this paper, we deeply compare six different convolutional network fusion architectures and analyse their adaptations, enabling a vanilla architecture to obtain detection performances comparable to the state-of-the-art results. Further, we discover that pedestrian detection confidences from color or thermal images are correlated with illumination conditions. With this in mind, we propose an Illumination-aware Faster R-CNN (IAF R-CNN). Specifically, an Illumination-aware Network is introduced to give an illumination measure of the input image. Then we adaptively merge color and thermal sub-networks via a gate function defined over the illumination value. The experimental results on KAIST Multispectral Pedestrian Benchmark validate the effectiveness of the proposed IAF R-CNN. [ABSTRACT FROM AUTHOR]

Highlights • This paper proposes a unified framework to perform classification and low-level modeling jointly. • Robustness is improved by considering a possibly badly labeled training set. • The proposed model allows a very rich interpretation of the modeled data structure. • Performance is assessed on synthetic and real data in the specific context of hyperspectral image interpretation. • The proposed model is generic enough to incorporate any kind of low-level modeling. Abstract Within a supervised classification framework, labeled data are used to learn classifier parameters. Prior to that, it is generally required to perform dimensionality reduction via feature extraction. These preprocessing steps have motivated numerous research works aiming at recovering latent variables in an unsupervised context. This paper proposes a unified framework to perform classification and low-level modeling jointly. The main objective is to use the estimated latent variables as features for classification and to incorporate simultaneously supervised information to help latent variable extraction. The proposed hierarchical Bayesian model is divided into three stages: a first low-level modeling stage to estimate latent variables, a second stage clustering these features into statistically homogeneous groups and a last classification stage exploiting the (possibly badly) labeled data. Performance of the model is assessed in the specific context of hyperspectral image interpretation, unifying two standard analysis techniques, namely unmixing and classification. [ABSTRACT FROM AUTHOR]

Highlights • Affine invariants constructed by traditional integer order moment are sensitive to noise. To construct invariants with lower (non-integer) order moment, we generalize the order of moment from integer to non-integer, and propose the Mellin polar coordinate moment (MPCM). • Method is provided for constructing affine invariants by any order MPCM. • Invariants constructed by lower real order MPCMs are more robust to noise than invariants constructed by traditional moments. Abstract The moment-based method is a fundamental approach to the extraction of affine invariants. However, only integer-order traditional moments can be used to construct affine invariants. No invariants can be constructed by moments with an order lower than 2. Consequently, the obtained invariants are sensitive to noise. In this paper, the moment order is generalized from integer to non-integer. However, the moment order cannot simply be generalized from integer to non-integer to achieve affine invariance. The difficulty of this generalization lies in the fact that the angular factor owing to shearing in the affine transform can hardly be eliminated for non-integer order moments. In order to address this problem, the Mellin polar coordinate moment (MPCM) is proposed, which is directly defined by a repeated integral. The angular factor can easily be eliminated by appropriately selecting a repeated integral. A method is provided for constructing affine invariants by means of MPCMs. The traditional affine moment invariants (AMIs) can be derived in terms of the proposed MPCM. Furthermore, affine invariants constructed with real-order (lower than 2) MPCMs can be derived using the proposed method. These invariants may be more robust to noise than AMIs. Several experiments were conducted to evaluate the proposed method performance. [ABSTRACT FROM AUTHOR]

Highlights • A novel strict and robust similarity estimator is proposed to guide the procedural model fitting. • A novel early rejection strategy is proposed to accelerate the procedural mode fitting. • The proposed method outperforms the state-of-the-art method in few-shot recognition. Abstract Procedural model fitting (PMF) is a generalization of classical model fitting and has numerous applications for computer vision and computer graphics. The task of PMF is to search a geometric model set for the model that is most similar to a set of data points. We propose a strict and robust similarity estimator for PMF to handle imperfect data. The proposed estimator is based on the error from model to data, while most other estimators are based on the error from data to model. We then use the proposed estimator to guide the cuckoo search algorithm to search for the most similar model. To accelerate the search process, we also propose a coarse-to-fine model dividing strategy to early reject dissimilar models. In this paper, the proposed PMF method is applied to fit building models on laser scanning data. It is also applied to fit character models on eighteen variants of imperfect MNIST data to achieve few-shot pattern recognition. In the 5-shot recognition, our method outperforms the state-of-the-art method on thirteen variants of the imperfect data. In particular, for one of the data corrupted by grid lines, our method obtains a high accuracy of 65%, whereas the state-of-the-art method only obtains an accuracy of 30%. [ABSTRACT FROM AUTHOR]

Comparative radiography is a forensic identification technique traditionally involving the manual comparison of ante-mortem and post-mortem radiographs, thus being time consuming and error prone. The main objective is to propose and validate a computer-aided comparative radiography paradigm based on the 3D bone scan-2D radiograph superimposition process of any bone or cavity. The proposal follows an image registration methodology to automatically search for the ante-mortem radiograph acquisition parameters from the forensic object’s silhouette considering occlusions. The underlying optimization problem is complex since a close initialization cannot be assumed and the image intensities are not reliable or not captured. Several experiments were performed to validate the method. First, we study its accuracy and robustness with synthetic images of clavicles, patellae and frontal sinuses. Second, we study how optimization performance and both variability and differences in the segmentation performed by human operators affect the identification using synthetic and real images of frontal sinuses. [ABSTRACT FROM AUTHOR]

Vote-boosting is a sequential ensemble learning method in which the individual classifiers are built on different weighted versions of the training data. To build a new classifier, the weight of each training instance is determined in terms of the degree of disagreement among the current ensemble predictions for that instance. For low class-label noise levels, especially when simple base learners are used, emphasis should be made on instances for which the disagreement rate is high. When more flexible classifiers are used and as the noise level increases, the emphasis on these uncertain instances should be reduced. In fact, at sufficiently high levels of class-label noise, the focus should be on instances on which the ensemble classifiers agree. The optimal type of emphasis can be automatically determined using cross-validation. An extensive empirical analysis using the beta distribution as emphasis function illustrates that vote-boosting is an effective method to generate ensembles that are both accurate and robust. [ABSTRACT FROM AUTHOR]

Visual tracking is still a challenging task as the objects suffer significant appearance changes, fast motion, and serious occlusion. In this paper, we propose an occlusion-aware part-based tracker for robust visual tracking. We first present a novel occlusion-aware part-based model based on correlation filters to integrate the global model and part-based model adaptively. It can effectively employ both the global and local information to improve the robustness of the tracker. Then we propose an integral pipeline aiming to the long-term tracking under the correlation filters, which achieves the state-of-the-art performance. In this tracking pipeline, we adopt separate translation and scale estimation. For translation estimation, we exploit and jointly learn the hierarchical features of deep Convolutional Neural Networks (CNNs) to locate the target center accurately. Then we learn an independent scale correlation filter to handle the scale variation. This design realizes scale adaptation of the target preferably, and reduces computational complexity efficiently. We further ameliorate the model update method by introducing the original reliable information. It greatly alleviates the error accumulation of the incorrect information and efficiently achieves long-term tracking. Extensive experimental results on several different challenging benchmark datasets show that our proposed tracker achieves outstanding performance against the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

For sparse representation or sparse coding based image classification, the dictionary, which is required to faithfully and robustly represent query images, plays an important role on its success. Learning dictionaries from the training data for sparse coding has shown state-of-the-art results in image classification and face recognition. However, for face recognition, conventional dictionary learning methods cannot well learn a reliable and robust dictionary due to suffering from the small-sample-size problem. The other significant issue is that current dictionary learning do not completely cover the important components of signal representation (e.g., commonality, particularity, and disturbance), which limit their performance. In order to solve the above issues, in this paper, we propose a novel robust, discriminative and comprehensive dictionary learning (RDCDL) method, in which a robust dictionary is learned from comprehensive training sample diversities generated by extracting and generating facial variations. Especially, to completely represent the commonality, particularity and disturbance, class-shared, class-specific and disturbance dictionary atoms are learned to represent the data from different classes. Discriminative regularizations on the dictionary and the representation coefficients are used to exploit discriminative information, which effectively improves the classification capability of the dictionary. The proposed RDCDL method is extensively evaluated on benchmark face image databases, and it shows superior performance to many state-of-the-art dictionary learning methods for face recognition. [ABSTRACT FROM AUTHOR]

This paper explores the problem of circle fitting for incomplete (partial arc) laser scanning point cloud data in the presence of outliers. In mobile laser scanning, data are commonly incomplete because of the orientation of the scanning unit to the surveying objects and the limited street-based positions. Also, multiple structures in the built environment often produce clustered outliers. To address these problems, this paper combines robust Principal Component Analysis (PCA) and robust regression with an efficient algebraic circle fitting method to develop two algorithms for circle fitting. Experimental efforts show that the proposed algorithms are statistically robust and can tolerate a high-percentage (exceeding 44%) of clustered outliers with insignificant error levels, while still achieving better shape recognition compared to existing competitive methods. For example, for a simulation of 1000 quarter circle datasets including 20% clustered outliers, RANSAC estimated the circle radius with a Mean Squared Error (MSE) of 172.10, whereas the proposed algorithms fit circles with an MSE of less than 0.42. The algorithms have potential in many areas including building information modeling, particle tracking, product quality control, arboreal assessment, and road asset monitoring. [ABSTRACT FROM AUTHOR]

This paper presents a robust regression approach for image binarization under significant background variations and observation noise. The work is motivated by the need of identifying foreground regions in noisy microscopic images or degraded document images, where significant background variations and observation noise make image binarization challenging. The proposed method first estimates the background of an input image, subtracts the estimated background from the input image, and performs a global thresholding operation to the subtracted outcome thus achieving the binary image of the foreground. A robust regression approach is proposed to estimate the background intensity surface with minimal effects of the foreground intensities and observation noise, and a global threshold selector is proposed on the basis of a model selection criterion in a sparse regression. The proposed approach is validated using 26 test images and the corresponding ground truths, and the outcomes are compared with those of nine existing image binarization methods. The approach is also combined with three morphological segmentation methods to show how the proposed approach can improve their image segmentation outcomes. [ABSTRACT FROM AUTHOR]

One of the major challenges in person Re-Identification (ReID) is the inconsistent visual appearance of a person. Current works on visual feature and distance metric learning have achieved significant achievements, but still suffer from the limited robustness to pose variations, viewpoint changes, etc ., and the high computational complexity. This makes person ReID among multiple cameras still challenging. This work is motivated to learn mid-level human attributes which are robust to visual appearance variations and could be used as efficient features for person matching. We propose a weakly supervised multi-type attribute learning framework which considers the contextual cues among attributes and progressively boosts the accuracy of attributes only using a limited number of labeled data. Specifically, this framework involves a three-stage training. A deep Convolutional Neural Network (dCNN) is first trained on an independent dataset labeled with attributes. Then it is fine-tuned on another dataset only labeled with person IDs using our defined triplet loss. Finally, the updated dCNN predicts attribute labels for the target dataset, which is combined with the independent dataset for the final round of fine-tuning. The predicted attributes, namely deep attributes exhibit promising generalization ability across different datasets. By directly using the deep attributes with simple Cosine distance, we have obtained competitive accuracy on four person ReID datasets. Experiments also show that a simple distance metric learning modular further boosts our method, making it outperform many recent works. [ABSTRACT FROM AUTHOR]

We survey unsupervised machine learning algorithms in the context of outlier detection. This task challenges state-of-the-art methods from a variety of research fields to applications including fraud detection, intrusion detection, medical diagnoses and data cleaning. The selected methods are benchmarked on publicly available datasets and novel industrial datasets. Each method is then submitted to extensive scalability, memory consumption and robustness tests in order to build a full overview of the algorithms’ characteristics. [ABSTRACT FROM AUTHOR]

Traditional depth estimation from stereo images is usually formulated as a patch-matching problem, which requires post-processing stages to impose smoothness and handle depth discontinuities and occlusions. While recent deep network approaches directly learn a regressor for the entire disparity map, they still suffer from large errors near the depth discontinuities. In this paper, we propose a novel method to refine the disparity maps generated by deep regression networks. Instead of relying on ad hoc post-processing, we learn a unified deep network model that predicts a confidence map and the disparity gradients from the learned feature representation in regression networks. We integrate the initial disparity estimation, the confidence map and the disparity gradients into a continuous Markov Random Field (MRF) for depth refinement, which is capable of representing rich surface structures. Our disparity MRF model can be solved via efficient global optimization in a closed form. We evaluate our approach on both synthetic and real-world datasets, and the results show it achieves the state-of-art performance and produces more structure-preserving disparity maps with smaller errors in the neighborhood of depth boundaries. [ABSTRACT FROM AUTHOR]

In this paper, we construct a least squares version of the recently proposed twin bounded support vector machine (TBSVM) for binary classification. As a valid classification tool, TBSVM attempts to seek two non-parallel planes that can be produced by solving a pair of quadratic programming problems (QPPs), but this is time-consuming. Here, we solve two systems of linear equations rather than two QPPs to avoid this deficiency. Furthermore, the distance in least squares TBSVM (LSTBSVM) is measured by L2-norm, but L1-norm distance is usually regarded as an alternative to L2-norm to improve model robustness in the presence of outliers. Inspired by the advantages of least squares twin support vector machine (LSTWSVM), TBSVM and L1-norm distance, we propose a LSTBSVM based on L1-norm distance metric for binary classification, termed as L1-LSTBSVM, which is specially designed for suppressing the negative effect of outliers and improving computational efficiency in large datasets. Then, we design a powerful iterative algorithm to solve the L1-norm optimal problems, and it is easy to implement and its convergence to an optimum solution is theoretically ensured. Finally, the feasibility and effectiveness of L1-LSTBSVM are validated by extensive experimental results on both UCI datasets and artificial datasets. [ABSTRACT FROM AUTHOR]

In word spotting literature, many approaches have considered word images as temporal signals that could be matched by classical Dynamic Time Warping algorithm. Consequently, DTW has been widely used as a on the shelf tool. However there exists many other improved versions of DTW, along with other robust sequence matching techniques. Very few of them have been studied extensively in the context of word spotting whereas it has been well explored in other application domains such as speech processing, data mining etc. The motivation of this paper is to investigate such area in order to extract significant and useful information for users of such techniques. More precisely, this paper has presented a comparative study of classical Dynamic Time Warping (DTW) technique and many of its improved modifications, as well as other sequence matching techniques in the context of word spotting, considering both theoretical properties as well as experimental ones. The experimental study is performed on historical documents, both handwritten and printed, at word or line segmentation level and with a limited or extended set of queries. The comparative analysis is showing that classical DTW remains a good choice when there is no segmentation problems for word extraction. Its constrained version (e.g. Itakura Parallelogram) seems better on handwritten data, as well as Hilbert transform also shows promising performances on handwritten and printed datasets. In case of printed data and low level features (pixel’s column based), the aggregation of features (e.g. Piecewise-DTW) seems also very important. Finally, when there are important word segmentation errors or when we are considering line segmentation level, Continuous Dynamic Programming (CDP) seems to be the best choice. [ABSTRACT FROM AUTHOR]

Person re-identification (re-id) aims to match pedestrians observed by disjoint camera views. It attracts increasing attention in computer vision due to its importance to surveillance systems. To combat the major challenge of cross-view visual variations, deep embedding approaches are proposed by learning a compact feature space from images such that the Euclidean distances correspond to their cross-view similarity metric. However, the global Euclidean distance cannot faithfully characterize the ideal similarity in a complex visual feature space because features of pedestrian images exhibit unknown distributions due to large variations in poses, illumination and occlusion. Moreover, intra-personal training samples within a local range which are robust to guide deep embedding against uncontrolled variations cannot be captured by a global Euclidean distance. In this paper, we study the problem of person re-id by proposing a novel sampling to mine suitable positives ( i.e., intra-class) within a local range to improve the deep embedding in the context of large intra-class variations. Our method is capable of learning a deep similarity metric adaptive to local sample structure by minimizing each sample’s local distances while propagating through the relationship between samples to attain the whole intra-class minimization. To this end, a novel objective function is proposed to jointly optimize similarity metric learning, local positive mining and robust deep feature embedding. This attains local discriminations by selecting local-ranged positive samples, and the learned features are robust to dramatic intra-class variations. Experiments on benchmarks show state-of-the-art results achieved by our method. [ABSTRACT FROM AUTHOR]

We introduce a fast Branch-and-Bound algorithm for optimal feature selection based on a U-curve assumption for the cost function. The U-curve assumption, which is based on the peaking phenomenon of the classification error, postulates that the cost over the chains of the Boolean lattice that represents the search space describes a U-shaped curve. The proposed algorithm is an improvement over the original algorithm for U-curve feature selection introduced recently. Extensive simulation experiments are carried out to assess the performance of the proposed algorithm (IUBB), comparing it to the original algorithm (UBB), as well as exhaustive search and Generalized Sequential Forward Search. The results show that the IUBB algorithm makes fewer evaluations and achieves better solutions under a fixed computational budget. We also show that the IUBB algorithm is robust with respect to violations of the U-curve assumption. We investigate the application of the IUBB algorithm in the design of imaging W -operators and in classification feature selection, using the average mean conditional entropy (MCE) as the cost function for the search. [ABSTRACT FROM AUTHOR]

Squared planar markers are a popular tool for fast, accurate and robust camera localization, but its use is frequently limited to a single marker, or at most, to a small set of them for which their relative pose is known beforehand. Mapping and localization from a large set of planar markers is yet a scarcely treated problem in favour of keypoint-based approaches. However, while keypoint detectors are not robust to rapid motion, large changes in viewpoint, or significant changes in appearance, fiducial markers can be robustly detected under a wider range of conditions. This paper proposes a novel method to simultaneously solve the problems of mapping and localization from a set of squared planar markers. First, a quiver of pairwise relative marker poses is created, from which an initial pose graph is obtained. The pose graph may contain small pairwise pose errors, that when propagated, leads to large errors. Thus, we distribute the rotational and translational error along the basis cycles of the graph so as to obtain a corrected pose graph. Finally, we perform a global pose optimization by minimizing the reprojection errors of the planar markers in all observed frames. The experiments conducted show that our method performs better than Structure from Motion and visual SLAM techniques. [ABSTRACT FROM AUTHOR]

Multidimensional Scaling (MDS) has been exploited to visualise the hidden structures among a set of entities in a reduced dimensional metric space. Here, we are interested in cases whenever the initial dissimilarity matrix is contaminated by outliers. It is well-known that the state-of-the-art algorithms for solving the MDS problem generate erroneous embeddings due to the distortion introduced by such outliers. To remedy this vulnerability, a unified framework for the solution of MDS problem is proposed, which resorts to half-quadratic optimization and employs potential functions of M -estimators in combination with ℓ 2,1 norm regularization. Two novel algorithms are derived. Their performance is assessed for various M -estimators against state-of-the-art MDS algorithms on four benchmark data sets. The numerical tests demonstrate that the proposed algorithms perform better than the competing alternatives. [ABSTRACT FROM AUTHOR]

In this paper, we propose a novel low-rank double dictionary learning (LRD 2 L) method for robust image classification tasks, in which the training and testing samples are both corrupted. Unlike traditional dictionary learning methods, LRD 2 L simultaneously learns three components from corrupted training data: 1) a low-rank class-specific sub-dictionary for each class to capture the most discriminative class-specific features of each class, 2) a low-rank class-shared dictionary which models the common patterns shared in the data of different classes, and 3) a sparse error term to model the noise in data. Through low-rank class-shared dictionary and noise term, the proposed method can effectively separate the corruptions and noise in training samples from creating low-rank class-specific sub-dictionaries, which are employed for correctly reconstructing and classifying testing images. Comparative experiments are conducted on three public available databases. Experimental results are encouraging, demonstrating the effectiveness of the proposed method and its superiority in performance over the state-of-the-art dictionary learning methods. [ABSTRACT FROM AUTHOR]

In fuzzy clustering, the fuzzy c-means (FCM) algorithm is the most commonly used clustering method. Various extensions of FCM had been proposed in the literature. However, the FCM algorithm and its extensions are usually affected by initializations and parameter selection with a number of clusters to be given a priori. Although there were some works to solve these problems in FCM, there is no work for FCM to be simultaneously robust to initializations and parameter selection under free of the fuzziness index without a given number of clusters. In this paper, we construct a robust learning-based FCM framework, called a robust-learning FCM (RL-FCM) algorithm, so that it becomes free of the fuzziness index m and initializations without parameter selection, and can also automatically find the best number of clusters. We first use entropy-type penalty terms for adjusting bias with free of the fuzziness index, and then create a robust learning-based schema for finding the best number of clusters. The computational complexity of the proposed RL-FCM algorithm is also analyzed. Comparisons between RL-FCM and other existing methods are made. Experimental results and comparisons actually demonstrate these good aspects of the proposed RL-FCM where it exhibits three robust characteristics: 1) robust to initializations with free of the fuzziness index, 2) robust to (without) parameter selection, and 3) robust to number of clusters (with unknown number of clusters). [ABSTRACT FROM AUTHOR]

The robust structural feature extraction and similarity measure play critical roles in person re-identification. This paper presents a novel algorithm named Multiple Metric Learning based on Bar-shape Descriptor ( MMLBD ) for person re-identification. Specifically, we first propose a new Multiple Bar-shape Descriptor that can take full account of the spatial correlation between the center points and their adjacent points on different directions. It captures further histogram features based on a novel color difference weight factors with an overlapping sliding window, which can depict the local variations and consistency in the whole image. The similarity and dissimilarity of samples are used to train the weight factor of features and an optimal subspace could be obtained at the same time. Next, we provide an effective multiple metric learning method fusing two-channel bar-shape structural features via the optimal similarity pairwise measure obtained by a dissimilarity matrix. This measure can fully mine the discriminative information and eliminate redundancy in the similar features, which make the MMLBD simple and effective. Finally, evaluation experiments on the i_LIDS, CAVIAR4REID and WARD data-sets are carried out, which compare the proposed MMLBD with the corresponding methods. Experimental results demonstrate that the MMLBD is more effective and robust against visual appearance variations. [ABSTRACT FROM AUTHOR]

Motivated by the success of powerful while expensive techniques to recognize words in a holistic way (Goel et al., 2013; Almazán et al., 2014; Jaderberg et al., 2016) object proposals techniques emerge as an alternative to the traditional text detectors. In this paper we introduce a novel object proposals method that is specifically designed for text. We rely on a similarity based region grouping algorithm that generates a hierarchy of word hypotheses. Over the nodes of this hierarchy it is possible to apply a holistic word recognition method in an efficient way. Our experiments demonstrate that the presented method is superior in its ability of producing good quality word proposals when compared with class-independent algorithms. We show impressive recall rates with a few thousand proposals in different standard benchmarks, including focused or incidental text datasets, and multi-language scenarios. Moreover, the combination of our object proposals with existing whole-word recognizers (Almazán et al., 2014; Jaderberg et al., 2016) shows competitive performance in end-to-end word spotting, and, in some benchmarks, outperforms previously published results. Concretely, in the challenging ICDAR2015 Incidental Text dataset, we overcome in more than 10% F -score the best-performing method in the last ICDAR Robust Reading Competition (Karatzas, 2015). Source code of the complete end-to-end system is available at https://github.com/lluisgomez/TextProposals . [ABSTRACT FROM AUTHOR]

In Smart and Autonomous Systems (SAS) with a discriminative tracking system, the general way to choose the training examples is to sample the image patches surrounding the tracking object at the current frame and label them manually. However, when the tracking result is not quite accurate, the classifier updates with misaligned examples that bring unexpected noise, which could result in drift. We propose a method of example selection based on multiple instance learning (MIL) using an iterative strategy. The idea is to remove the worst example that has little contributions to the likelihood of bags in the positive bag for each round. The proposed system is built on FCT, a novel tracking algorithm that adopts compressive sensing for image representation. The qualitative and quantitative experimental results show the advantage of the proposed tracker in efficiency and accuracy to current state-of-the-art trackers when objects suffer from deformation and rotation. [ABSTRACT FROM AUTHOR]

This paper presents a compact and efficient yet powerful binary framework based on image gradients for robust facial representation. It is termed as Binary Gradient Patterns (BGP). To discover underlying local structures in the gradient domain, image gradients are computed from multiple directions and encoded into a set of binary strings. Certain types of these binary strings have meaningful local structures and textures, as they detect micro oriented edges and retain strong local orientation, thus enabling great discrimination. Face representations by these structural BGP histograms exhibit profound robustness against various facial image variations, in particular illumination. The binary strategy realized by local correlations substantially simplifies the computational complexity and achieves extremely efficient processing with only 0.0032 s in Matlab for a typical image. Furthermore, the discrimination power of the BGP has been enhanced on a set of orientations of the image-gradient magnitudes. Extensive experimental results on various benchmarks demonstrate that the BGP-based representations significantly improve over the existing local descriptors and state-of-the-art methods in the terms of discrimination, robustness and complexity and in many cases the improvements are substantial. Combining with the deep networks, the proposed descriptors can further improve the performance of the deep networks on real-world datasets. Matlab codes for the BGP-based descriptors are available at: http://www.eee.manchester.ac.uk/our-research/research-groups/sisp/research-areas/vip/software/ . [ABSTRACT FROM AUTHOR]

Feature noise, namely noise on inputs is a long-standing plague to support vector machine(SVM). Conventional SVM with the hinge loss( C -SVM) is sparse but sensitive to feature noise. Instead, the pinball loss SVM( pin -SVM) enjoys noise robustness but loses the sparsity completely. To bridge the gap between C -SVM and pin -SVM, we propose the truncated pinball loss SVM( p i n ¯ -SVM) in this paper. It provides a flexible framework of trade-off between sparsity and feature noise insensitivity. Theoretical properties including Bayes rule, misclassification error bound, sparsity, and noise insensitivity are discussed in depth. To train p i n ¯ -SVM, the concave-convex procedure(CCCP) is used to handle non-convexity and the decomposition method is used to deal with the subproblem of each CCCP iteration. Accordingly, we modify the popular solver LIBSVM to conduct experiments and numerical results validate the properties of p i n ¯ -SVM on the synthetic and real-world data sets. [ABSTRACT FROM AUTHOR]

Two-dimensional principal component analysis (2DPCA) employs the squared F-norm as distance metric for feature extraction and is widely used in the field of pattern analysis and recognition, especially face image analysis. But it is sensitive to the presence of outliers due to the fact that squared F-norm remarkably enlarges the role of outliers in the criterion function. To handle this problem, we propose a robust formulation for 2DPCA, namely optimal mean 2DPCA with F-norm minimization (OMF-2DPCA). In OMF-2DPCA, distance in spatial dimensions (attribute dimensions) is measured in F-norm, while the summation over different data points uses 1-norm. Moreover, we center the data using the optimized mean rather than the fixed mean. This helps further improve robustness of our method. To solve OMF-2DPCA, we propose a fast iterative algorithm, which has a closed-form solution in each iteration. Experimental results on face image databases illustrate its effectiveness and advantages. [ABSTRACT FROM AUTHOR]

Sparse representation based face recognition (SRC) has been paid much attention in recent years. Representative algorithms are deformable sparse-representation based classification (DSRC) and shape-constrained texture matching, which focus on misalignment and shape change respectively. Although these algorithms obtain improved accuracy and robustness, their efficiency is not satisfied particularly when applied in a large-scale system. The main difficulty is the expensive calculation in aligning gallery images and the probe image. To solve these problems, this paper proposes a fast alignment strategy for sparse representation based algorithms. The key idea is to pre-compute the most expensive operation, Hessian matrix, which was needed to be calculated in each iteration. Subsequently, with help of the proposed fast alignment strategy, two algorithms, deformable SRC and shape-constrained texture matching, are extended to their fast versions, i.e., fast deformable SRC and fast shape-constrained texture matching. Experimental evaluations have been conducted on the public datasets such as Multi-PIE, FERET, and Cohn-Kanade. We demonstrate that the proposed alignment strategy greatly improves the efficiency of original algorithms without losing accuracy and robustness. [ABSTRACT FROM AUTHOR]

Tackling illumination variation is a major problem and it is also an important challenge for practical face recognition systems. Some related methods consider that lighting intensity components mainly lie in large-scale features, and they use a lot of image decomposition techniques to extract the small-scale illumination-invariant features and remove the large-scale features from original face images. However, it argues that the large-scale features contain a lot useful information which can be further extracted, and the small-scale illumination-invariant features are not robust enough due to they contain some detrimental features (noise, etc.). In this paper, we propose a discriminative multi-layer illumination-robust feature extraction (DMI) model to address this problem. First, we decompose the large-scale features into multi-layer small-scale illumination-robust features as a linear combination, and then a weight is assigned to each layer to adjust its importance and influence. The idea is to take full advantage of these useful information in large-scale features for face recognition. Second, we learn a discriminant filter to improve the robustness and statistical discriminative ability of the reconstructed illumination-robust face for face recognition under poor lighting conditions. Extensive experiments on three benchmark face databases and a video image database show that DMI performs better than the related methods, especially in difficult lighting conditions. [ABSTRACT FROM AUTHOR]

Due to its ability to eliminate the visual ambiguities in single-shot algorithms, video-based person re-identification has received an increasing focus in computer vision. Visual ambiguities caused by variations in view angle, lighting, and occlusions make the re-identification problem extremely challenging. To overcome the ambiguities, most previous approaches often extract robust feature representations or learn a sophisticated feature transformation. However, most of these approaches ignore the effect of the impostors arising from annotation or tracking process. In this case, impostors are regarded as genuine and applied in training process, leading to the model drift problem. In order to reduce the risk of model drifting, we propose to automatically discover impostors in a multiple instance metric learning framework. Specifically, we propose a k NN based confidence score to evaluate how much an impostor invades the interested target and utilize it as a prior in the framework. In the meanwhile, we integrate an impostor rejection mechanism in the multiple instance metric learning framework to automatically discover impostors, and learn the semantical similarity metrics with the refined training set. Experiments show that the proposed system performs favorably against the state-of-the-art algorithms on two challenging datasets (iLIDS-VID and PRID 2011). We have improved the rank 1 recognition rate on iLIDS-VID and PRID 2011 dataset by 1.0% and 1.2%, respectively. [ABSTRACT FROM AUTHOR]

Most multi-task learning based trackers adopt similar task definition by assuming that all tasks share a common feature set, which can’t cover the real situation well. In this paper, we define the subtasks from the novel perspective, and develop a structured and consistent multi-layer multi-subtask tracker with graph regularization. The tracking task is completed by the collaboration of multi-layer subtasks. Different subtasks correspond to the tracking of different parts in the target area. The correspondences of the subtasks among the adjacent frames are consistent and smooth. The proposed model introduces hyper-graph regularizer to preserve the global and local intrinsic geometrical structures among and inside target candidates or trained samples, and decomposes the representative matrix of the subtasks into two components: low-rank property captures the subtask relationship, group-sparse property identifies the outlier subtasks. Moreover, a collaborate metric scheme is developed to find the best candidate, by concerning both discrimination reliability and representation accuracy. We show that the proposed multi-layer multi-subtask learning based tracker is a general model, which accommodates most existing multi-task trackers with the respective merits. Encouraging experimental results on a large set of public video sequences justify the effectiveness and robustness of the proposed tracker, and achieve comparable performance against many state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Accurate and reliable optic disc (OD) segmentation is important for retinal image analysis and retinal disease screening. This paper presents a novel method to automatically segment OD in fundus images based on variational model with multiple energies. Firstly, a sparse coding based technique is designed to localize the OD center, based on which an initial boundary curve is then estimated by a circular Hough transform. Next, OD segmentation is regarded as an energy minimization problem, and a variational model integrating three energy terms is proposed to evolve the curve to the OD boundary. In the proposed model, the first term, named phase-based boundary energy, is designed to attract the evolution curve to the OD boundary, even the one with low contrast; the second term, named PCA-based shape energy, constraints the evolution curve to a common OD shape, which can suppress the negative effect of bright interferences, e.g., the bright lesions and myelinated nerve fibers, in OD segmentation; the last one is the region energy, which drives the evolution curve to the boundary of the homogeneous regions and hence improve the robustness of the model to the noises boundary and the initial position of evolution curve. The proposed OD segmentation method is evaluated on three public available databases, i.e., the MESSIDOR, ONHSD and DRIONS databases, and the experimental results demonstrate that the proposed method outperforms the state-of-the art techniques. [ABSTRACT FROM AUTHOR]

Category-specific object segmentation has been a long-standing research topic in pattern recognition. This paper presents an unsupervised discriminant shape (UDS) to address category-specific object segmentation by incorporating the proposed shape prior into an intuitive energy minimization framework. Recently, based on the region proposal methods, deep Convolutional Neural Networks (CNNs) provide access to candidate segments in categories of interest from images. However, the segments obtained from bottom-up proposals tend to undershoot or overshoot objects and are easily classified into one specific class. To address this problem, we propose an unsupervised discriminant projection based clustering algorithm (UDC) to obtain more precise shape prior to guide the segmentation, and the class-specific proposals are clustered based on their projections onto the discriminant projection direction. Based on the set of proposals, we then obtain the prior information of foreground UDS with an easy voting scheme. The derived UDS prior is finally utilized in the subsequent energy minimizing formulation based figure-ground segmentation. We conduct extensive and comprehensive evaluations on the MSRC, Object Discovery, Fashionista and PASCAL-S datasets, demonstrating the effectiveness and robustness of the UDS based segmentation. [ABSTRACT FROM AUTHOR]

We introduce the implicitly constrained least squares (ICLS) classifier, a novel semi-supervised version of the least squares classifier. This classifier minimizes the squared loss on the labeled data among the set of parameters implied by all possible labelings of the unlabeled data. Unlike other discriminative semi-supervised methods, this approach does not introduce explicit additional assumptions into the objective function, but leverages implicit assumptions already present in the choice of the supervised least squares classifier. This method can be formulated as a quadratic programming problem and its solution can be found using a simple gradient descent procedure. We prove that, in a limited 1-dimensional setting, this approach never leads to performance worse than the supervised classifier. Experimental results show that also in the general multidimensional case performance improvements can be expected, both in terms of the squared loss that is intrinsic to the classifier and in terms of the expected classification error. [ABSTRACT FROM AUTHOR]

In this paper, a novel active contours method, which combines with the Student's-t mixture model via Expectaton-Maximizaton (EM) algorithm, is proposed to segment complicated two-phase images. Firstly, we rewrite the cost function and derive a novel updating of level set function based on probabilistic principles. Secondly, we put forward two novel geometric priors from the level-set-based curve evolution; and both of them have advantages, the suitable one is selected by personalized need to obtain level set function in EM framework with the aim of reducing the computational cost. Therefore, the level set function is derived from latent variables and served as a feedback to the estimation of the latent variables in next iteration. Finally, in order to enhance the robustness to the outliers, Student's-t mixture model with heavy tail has been applied in our algorithm. Experimental results obtained by employing the proposed method on many synthetic, medical and real-world images to demonstrate its robustness, accuracy and effectiveness. [ABSTRACT FROM AUTHOR]

In this work we address the problem of real-time dynamic medical (MRI and X-Ray CT) image reconstruction from parsimonious samples (Fourier frequency space for MRI and sinogram/tomographic projections for CT). Today the de facto standard for such reconstruction is compressed sensing (CS). CS produces high quality images (with minimal perceptual loss); but such reconstructions are time consuming, requiring solving a complex optimization problem. In this work we propose to ‘learn’ the reconstruction from training samples using an autoencoder. Our work is based on the universal function approximation capacity of neural networks. The training time for the autoencoder is large, but is offline and hence does not affect performance during operation. During testing/operation, our method requires only a few matrix vector products and hence is significantly faster than CS based methods. In fact, for MRI it is fast enough for real-time reconstruction (the images are reconstructed as fast as they are acquired) with only slight degradation of image quality; for CT our reconstruction speed is slightly slower than required for real-time reconstruction. However, in order to make the autoencoder suitable for our problem, we depart from the standard Euclidean norm cost function of autoencoders and use a robust l 1 -norm instead. The ensuing problem is solved using the Split Bregman method. [ABSTRACT FROM AUTHOR]

Recently, multi-atlas patch-based label fusion has achieved many successes in medical imaging area. The basic assumption in the current state-of-the-art approaches is that the image patch at the target image point can be represented by a patch dictionary consisting of atlas patches from registered atlas images. Therefore, the label at the target image point can be determined by fusing labels of atlas image patches with similar anatomical structures. However, such assumption on image patch representation does not always hold in label fusion since (1) the image content within the patch may be corrupted due to noise and artifact; and (2) the distribution of morphometric patterns among atlas patches might be unbalanced such that the majority patterns can dominate label fusion result over other minority patterns. The violation of the above basic assumptions could significantly undermine the label fusion accuracy. To overcome these issues, we first consider forming label-specific group for the atlas patches with the same label. Then, we alter the conventional flat and shallow dictionary to a deep multi-layer structure, where the top layer ( label-specific dictionaries ) consists of groups of representative atlas patches and the subsequent layers ( residual dictionaries ) hierarchically encode the patchwise residual information in different scales. Thus, the label fusion follows the representation consensus across representative dictionaries. However, the representation of target patch in each group is iteratively optimized by using the representative atlas patches in each label-specific dictionary exclusively to match the principal patterns and also using all residual patterns across groups collaboratively to overcome the issue that some groups might be absent of certain variation patterns presented in the target image patch. Promising segmentation results have been achieved in labeling hippocampus on ADNI dataset, as well as basal ganglia and brainstem structures, compared to other counterpart label fusion methods. [ABSTRACT FROM AUTHOR]

Recognition of texts in scenes is one of the most important tasks in many computer vision applications. Though different scene text recognition techniques have been developed, scene text recognition under a generic condition is still a very open and challenging research problem. One major factor that defers the advance in this research area is character touching, where many characters in scene images are heavily touched with each other and cannot be segmented for recognition. In this paper, we proposed a novel scene text recognition technique that performs word level recognition without character segmentation. Our proposed technique has three advantages. First it converts each word image into a sequential signal for the scene text recognition. Second, it adapts the recurrent neural network (RNN) with Long Short Term Memory (LSTM), the technique that has been widely used for handwriting recognition in recent years. Third, by integrating multiple RNNs, an accurate recognition system is developed which is capable of recognizing scene texts including those heavily touched ones without character segmentation. Extensive experiments have been conducted over a number of datasets including several ICDAR Robust Reading datasets and Google Street View dataset. Experiments show that the proposed technique is capable of recognizing texts in scenes accurately. [ABSTRACT FROM AUTHOR]

This paper presents a novel noise robust edge detector based upon the automatic anisotropic Gaussian kernels (ANGKs), which also addresses the current problem that the seminal Canny edge detector may miss some obvious crossing edge details. Firstly, automatic ANGKs are designed according to the noise suppression, edge resolution and localization precision, which also conciliate the conflict between them. Secondly, reasons why cross-edge points are missing from Canny detector results using isotropic Gaussian kernel are analyzed. Thirdly, the automatic ANGKs are used to smooth image and a revised edge extraction method is used to extract edges. Finally, the aggregate test receiver-operating-characteristic (ROC) curves and Pratt's Figure of Merit (FOM) are used to evaluate the proposed detector against state-of-the-art edge detectors. The experiment results show that the proposed algorithm can obtain better performance for noise-free and noisy images. [ABSTRACT FROM AUTHOR]

The support vector machine ( SVM ) is a popular classifier in machine learning, but it is not robust to outliers. In this paper, based on the Correntropy induced loss function, we propose the rescaled hinge loss function which is a monotonic, bounded and nonconvex loss that is robust to outliers. We further show that the hinge loss is a special case of the proposed rescaled hinge loss. Then, we develop a new robust SVM based on the rescaled hinge loss. After using the half-quadratic optimization method, we find that the new robust SVM is equivalent to an iterative weighted SVM, which can help explain the robustness of iterative weighted SVM from a loss function perspective. Experimental results confirm that the new robust SVM not only performs better than SVM and the existing robust SVMs on the datasets that have outliers, but also presents better sparseness than SVM. [ABSTRACT FROM AUTHOR]

Single sample per person (SSPP) recognition is one of the most challenging problems in face recognition (FR) due to the lack of information to predict the variations in the query sample. To address this problem, we propose in this paper a novel face recognition algorithm based on a robust collaborative representation (CR) and probabilistic graph model, which is called Collaborative Probabilistic Labels (CPL). First, by utilizing label propagation, we construct probabilistic labels for the samples in the generic training set corresponding to those in the gallery set, thus the discriminative information of the unlabeled data can be effectively explored in our method. Then, the adaptive variation type for a given test sample is automatically estimated. Finally, we propose a novel reconstruction-based classifier for the test sample with its corresponding adaptive dictionary and probabilistic labels. The proposed probabilistic graph based model is adaptively robust to various variations in face images, including illumination, expression, occlusion, pose, etc., and is able to reduce required training images to one sample per class. Experimental results on five widely used face databases are presented to demonstrate the efficacy of the proposed approach. [ABSTRACT FROM AUTHOR]