Search

Feature selection is an important approach for reducing the dimension of high-dimensional data. In recent years, many feature selection algorithms have been proposed, but most of them only exploit information from the data space. They often neglect useful information contained in the feature space, and do not make full use of the characteristics of the data. To overcome this problem, this paper proposes a new unsupervised feature selection algorithm, called non-negative spectral learning and sparse regression-based dual-graph regularized feature selection (NSSRD). NSSRD is based on the feature selection framework of joint embedding learning and sparse regression, but extends this framework by introducing the feature graph. By using low dimensional embedding learning in both data space and feature space, NSSRD simultaneously exploits the geometric information of both spaces. Second, the algorithm uses non-negative constraints to constrain the low-dimensional embedding matrix of both feature space and data space, ensuring that the elements in the matrix are non-negative. Third, NSSRD unifies the embedding matrix of the feature space and the sparse transformation matrix. To ensure the sparsity of the feature array, the sparse transformation matrix is constrained using the ${L_{2,1}}$ -norm. Thus feature selection can obtain accurate discriminative information from these matrices. Finally, NSSRD uses an iterative and alternative updating rule to optimize the objective function, enabling it to select the representative features more quickly and efficiently. This paper explains the objective function, the iterative updating rules and a proof of convergence. Experimental results show that NSSRD is significantly more effective than several other feature selection algorithms from the literature, on a variety of test data.

In this paper, a robust point correspondence algorithm is proposed to address the problems with SIFT-like methods for the optical satellite image registration. SIFT-like methods involve two issues that are rarely noticed: non-orientation selectivity in feature detection and feature redundancy in feature description. The novelty of the proposed approach is that the advantages of biologically motivated methods are adopted to resolve above two problems. Firstly, by using a 2D Gabor filter bank to model the visual cognitive computational model, intuitive and robust keypoints are detected. The proposed detector can capture salient visual properties such as the orientation and spatial frequency selectivity. Secondly, multi-characteristic scales of the keypoints are selected based on the Gabor kernel function, and then multi-feature descriptors with high discriminating power are defined. By using the proposed detector and descriptor, the feature redundancy can translate into benefits. Finally, a feature matching strategy for multi-feature descriptors is designed, to improve the reliability of feature matching. Evaluation criteria of 1-precision, RMSE and visual inspection of the matched pairs are used to demonstrate the superior performance of the proposed algorithm on optical satellite image registration.

Although many feature-based registration methods have been proposed, automatic image registration is still a challenging task due to the influence of various conditions and uncertain difficulties for remote sensing images. In this letter, a novel image registration method, including two types of feature detectors and a region boundary constraint strategy for matching, is proposed. Two types of features detected by scale-invariant feature transform and Harris operators have advantages of keeping different structural information in the image and increasing the number of keypoints for later matching. Afterward, a region boundary constraint strategy based on the image sketch map is utilized in matching step. This strategy restricts the detected two types of features in their respective structural region and nonstructural region to reduce the incorrect correspondences. Experimental results demonstrate the superiority of our proposed registration algorithm compared with other research works in terms of correct matching number and aligning accuracy.

In the field of big data applications, image information is widely used. The value density of information utilization in big data is very low, and how to extract useful information quickly is very important. So we should transform the unstructured image data source into a form that can be analyzed. In this paper, we proposed a fast image retrieval method which designed for big data. First of all, the feature extraction method is necessary and the feature vectors can be obtained for every image. Then, it is the most important step for us to encode the image feature vectors and make them into database, which can optimize the feature structure. Finally, the corresponding similarity matching is used to determined the retrieval results. There are three main contributions for image retrieval in this paper. New feature extraction method, reasonable elements ranking, and appropriate distance metric can improve the algorithm performance. Experiments show that our method has a great improvement in the effective performance of feature extraction and can also get better search matching results.

This study considers the insufficiency of traditional monitoring methods to eliminate dynamics, and proposes a novel online feature reordering- and feature selection-based slow feature analysis (SFA) algorithm. The SFA algorithm explores the process dynamics from the view of inner variation of data to extract the slowly varying features. The extracted SFs are considered as the representations of steady- and dynamic-state processes. Online feature reordering and feature selection strategies maximize online fault information and can be used to perform fault detection operation. The proposed method is applied to two simulated processes. Monitoring results show that the proposed method has better monitoring results than those of traditional methods.

We propose a novel feature extraction approach for 3D facial expression recognition by incorporating non-rigid registration in face-model-free analysis, which in turn makes feasible data-driven, i.e., feature-model-free recognition of expressions. The resulting simplicity of feature representation is due to the fact that facial information is adapted to the input faces via shape model-free dense registration, and this provides a dynamic feature extraction mechanism. This approach eliminates the necessity of complex feature representations as required in the case of static feature extraction methods, where the complexity arises from the necessity to model the local context; higher degree of complexity persists in deep feature hierarchies enabled by end-to-end learning on large-scale datasets. Face-model-free recognition implies independence from limitations and biases due to committed face models, bypassing complications of model fitting, and avoiding the burden of manual model construction. We show via information gain maps that non-rigid registration enables extraction of highly informative features, as it provides invariance to local-shifts due to physiognomy (subject invariance) and residual pose misalignments; in addition, it allows estimation of local correspondences of expressions. To maximize the recognition rate, we use the strategy of employing a rich but computationally manageable set of local correspondence structures, and to this effect we propose a framework to optimally select multiple registration references. Our features are re-sampled surface curvature values at individual coordinates which are chosen per expression-class and per reference pair. We show the superior performance of our novel dynamic feature extraction approach on three distinct recognition problems, namely, action unit detection, basic expression recognition, and emotion dimension recognition.

Feature points are effective for wide range of computer vision applications. In the last two decades, a large number of feature point detection and description algorithms have been proposed. All these algorithms are implemented differently but have a common objective to detect and describe feature points invariant to scale, rotation, intensity, and affine variations. Several comparative studies of feature points have been reported in literature. These studies are either application specific or deal with common type of transformations and deformations. Additionally, they primarily focus on evaluation of feature points on gray scale or RGB images. In contrast, this paper presents a comparison of feature points on multisensor images, which possess non linear intensity changes. The objective is to identify robust feature points for image to image matching tasks on multisensor images. Six well known feature point detector and seventeen popular descriptor algorithms are compared. Experimental results obtained on four different image datasets show that the combination of Harris detector with BRIEF descriptor outperforms all other detector-descriptor combinations on multisensor images.

FCTFS works in both autonomous and user guided mode.The defined taxonomy helps in arriving at optimal number of good quality clusters.Feature elimination due to irrelevance and redundancy is clearly isolated.It is faster than traditional search based methods.Yields superior results compared to some state of the art methods over 24 data sets. Feature subset selection is basically an optimization problem for choosing the most important features from various alternatives in order to facilitate classification or mining problems. Though lots of algorithms have been developed so far, none is considered to be the best for all situations and researchers are still trying to come up with better solutions. In this work, a flexible and user-guided feature subset selection algorithm, named as FCTFS (Feature Cluster Taxonomy based Feature Selection) has been proposed for selecting suitable feature subset from a large feature set. The proposed algorithm falls under the genre of clustering based feature selection techniques in which features are initially clustered according to their intrinsic characteristics following the filter approach. In the second step the most suitable feature is selected from each cluster to form the final subset following a wrapper approach. The two stage hybrid process lowers the computational cost of subset selection, especially for large feature data sets. One of the main novelty of the proposed approach lies in the process of determining optimal number of feature clusters. Unlike currently available methods, which mostly employ a trial and error approach, the proposed method characterises and quantifies the feature clusters according to the quality of the features inside the clusters and defines a taxonomy of the feature clusters. The selection of individual features from a feature cluster can be done judiciously considering both the relevancy and redundancy according to users intention and requirement. The algorithm has been verified by simulation experiments with different bench mark data set containing features ranging from 10 to more than 800 and compared with other currently used feature selection algorithms. The simulation results prove the superiority of our proposal in terms of model performance, flexibility of use in practical problems and extendibility to large feature sets. Though the current proposal is verified in the domain of unsupervised classification, it can be easily used in case of supervised classification.

Figure-ground segmentation is the process of separating regions of interest from unimportant background. One challenge is to segment images with high variations (e.g. containing a cluttered background), which requires effective feature sets to capture the distinguishing information between objects and backgrounds. Feature selection is necessary to remove noisy/redundant features from those extracted by image descriptors. As a powerful search algorithm, genetic programming (GP) is employed for the first time to build feature selection methods that aims to improve the segmentation performance of standard classification techniques. Both single-objective and multi-objective GP techniques are investigated, based on which three novel feature selection methods are proposed. Specifically, one method is single-objective, called PGP-FS (parsimony GP feature selection); while the other two are multi-objective, named nondominated sorting GP feature selection (NSGP-FS) and strength Pareto GP feature selection (SPGP-FS). The feature subsets produced by the three proposed methods, two standard sequential selection algorithms, and the original feature set are tested via standard classification algorithms on two datasets with high variations (the Weizmann and Pascal datasets). The results show that the two multi-objective methods (NSGP-FS and SPGP-FS) can produce feature subsets that lead to solutions achieving better segmentation performance with lower numbers of features than the sequential algorithms and the original feature set based on standard classifiers for given segmentation tasks. In contrast, PGP-FS produces results that are not consistent for different classifiers. This indicates that the proposed multi-objective methods can help standard classifiers improve the segmentation performance while reducing the processing time. Moreover, compared with SPGP-FS, NSGP-FS is equally capable of producing effective feature subsets, yet is better at keeping diverse solutions.

Feature subset selection with the aim of reducing dependency of feature selection techniques and obtaining a high-quality minimal feature subset from a real-world domain is the main task of this research. For this end, firstly, two types of feature representation are presented for feature sets, namely unigram-based and part-of-speech based feature sets. Secondly, five methods of feature ranking are employed for creating feature vectors. Finally, we propose two methods for the integration feature vectors and feature subsets. An ordinal-based integration of different feature vectors (OIFV) is proposed in order to obtain a new feature vector. The new feature vector depends on the order of features in the old vectors. A frequency-based integration of different feature subsets (FIFS) with most effective features, which are obtained from a hybrid filter and wrapper methods in the feature selection task, is then proposed. In addition, four well-known text classification algorithms are employed as classifiers in the wrapper method for the selection of the feature subsets. A wide range of comparative experiments on five widely-used datasets in sentiment analysis were carried out. The experiments demonstrate that proposed methods can effectively improve the performance of sentiment classification. These results also show that proposed part-of-speech patterns are more effective in their classification accuracy compared to unigram-based features.

It is important that users can progressively simplify a detailed CAD model and create a simplified one for use in various applications. For this purpose, many simplification methods have been proposed, and it is known that feature-based simplification methods are superior for engineering applications. The previous feature-based simplification methods assumed that a feature-based CAD model has a single-branch feature tree, and the type of each feature is either additive or subtractive. However, a feature-based CAD model is often represented as a multi-branch feature tree. Furthermore, the type of a feature may be both additive and subtractive simultaneously. To solve these problems, we present an extended feature-based simplification method. In the proposed method, the modeling sequence of the features from a multi-branch feature tree is determined using the feature dependency graph. The importance of each feature is evaluated, and the features are rearranged by their importance using the proposed extended feature rearrangement algorithm. The feature-based CAD model is finally simplified in a manner where less important features are suppressed until the CAD model reaches the desired level of detail. To demonstrate the proposed method, a prototype system was implemented, and experiments with test cases were performed.

High dimensional feature extraction using MLBP maps is proposed.Each bit-plane decomposed from the MLBP has local-structure of the face image.Orthogonal LDA is applied to the high dimensional feature vector. Transforming an original image into a high-dimensional (HD) feature has been proven to be effective in classifying images. This paper presents a novel feature extraction method utilizing the HD feature space to improve the discriminative ability for face recognition. We observed that the local binary pattern can be decomposed into bit-planes, each of which has scale-specific directional information of the face image. Each bit-plane not only has the inherent local-structure of the face image but also has an illumination-robust characteristic. By concatenating all the decomposed bit-planes, we generate an HD feature vector with an improved discriminative ability. To reduce the computational complexity while preserving the incorporated local structural information, a supervised dimension reduction method, the orthogonal linear discriminant analysis, is applied to the HD feature vector. Extensive experimental results show that existing classifiers with the proposed feature outperform those with other conventional features under various illumination, pose, and expression variations.

The curse of dimensionality is based on the fact that high dimensional data is often difficult to work with. A large number of features can increase the noise of the data and thus the error of a learning algorithm. Feature selection is a solution for such problems where there is a need to reduce the data dimensionality. Different feature selection algorithms may yield feature subsets that can be considered local optima in the space of feature subsets. Ensemble feature selection combines independent feature subsets and might give a better approximation to the optimal subset of features. We propose an ensemble feature selection approach based on feature selectors’ reliability assessment. It aims at providing a unique and stable feature selection without ignoring the predictive accuracy aspect. A classification algorithm is used as an evaluator to assign a confidence to features selected by ensemble members based on their associated classification performance. We compare our proposed approach to several existing techniques and to individual feature selection algorithms. Results show that our approach often improves classification performance and feature selection stability for high dimensional data sets.

Feature point matching is the process of finding an optimal spatial transformation that aligns two arbitrary sets of feature points. It is one of the most fundamental problems in the computer vision domain and is frequently used in object recognition, image registration, camera self-calibration, and so on. Critical to most feature point matching techniques is the determination of correspondence between spatially localized feature points within each image. Moreover, there can be many feature points in either set that have no counterparts in the other. A robust and effective method for feature point matching is thus required and is still a challenge. In this work, an artificial bee colony (ABC) with a normalized cross-correlation (NCC) algorithm called “ABC-NCC” for feature point matching is presented. In this proposed method, both the size and the orientation of the correlation window used for calculating the NCC are determined according to the scale and the rotation direction of the interest points, which are optimized by the ABC algorithm. Experimental results obtained by our method show that the proposed approach works well for feature point matching and outperforms existing algorithms.

In this paper, we propose a new image registration method for synthetic aperture radar (SAR) image with multiscale image patch features, in which the sparse representation technique is exploited. Considering the influence of speckle noise on feature extraction, in the proposed method, a spatial correlation strategy based on stationary wavelet transform is adopted to select the reliable feature points from the initial scale invariant feature transform keypoints in the reference image. By introducing multiscale image patch, a new feature descriptor is further designed to describe the attribute domain of feature points for higher discrimination. The corresponding points in the sensed image are established based on the minimum discrepancy criterion calculated by the sparse representation technique. Moreover, the local geometric consistency among a feature point and its nearest neighbor points is employed to remove the mismatches from the tentative matches. Twenty-two pairs of SAR images acquired under various conditions are utilized to validate the effectiveness of the proposed method. Compared with the traditional SAR image registration methods, the results show that the proposed method is competent to improve the registration performance substantially.

Dimensionality reduction is an essential pre-processing technique in many of the data analysis tasks. Popular approaches for dimensionality reduction are Feature Selection (FS) and Feature Extraction (FE). Till now, these approaches are often studied separately or independently so that the final result contains either original or transformed features. In our work, we propose to bridge these two approaches with the aim of finding reduced feature set to contain both kinds (original as well as transformed) of features. A new framework, called Minimum Projection error Minimum Redundancy (MPeMR), is introduced to obtain this result while maintaining orthogonality property among selected original and linear combinations of features. A unified iterative algorithm, for both supervised and unsupervised cases, is also developed under this framework. For each case, the performance of the proposed algorithm is successfully compared with the state-of-the-art methods on real-life data sets.

High-dimension of feature space in text classification is a major problem of it. Feature selection is an effective method for feature reduction. A multiple centrifuge models based feature selection method is put forward in the view of the hypothesis that the same documents have core feature set in the text classification and the classes of the same high-frequency feature words of document have affinity. The proposed feature selection algorithm made a lot of innovation ideas in the field of feature reduction which improve the values of the low-frequency features in classification meanwhile ensuring the classification effect. The experiments in the Reuters-21578 corpus show that this method has better classification effect, and effectively improves the utilization of medium or low frequency features which have strong classification ability.

Feature representation is a classic problem in the machine learning community due to the fact that different representations can entangle and hide more or less the different explanatory factors of variation behind the raw data. Especially for scene classification, its performance generally depends on the discriminative power of feature representation. Recently, unsupervised feature learning attracts tremendous attention because of its ability to learn feature representation automatically. However, reliable performance of feature representations by unsupervised learning always requires a large number of features and complex framework of mid-level feature representation. To alleviate such drawbacks, this paper presents a new framework of mid-level feature representation, which does not need learn many convolutional features during the unsupervised feature learning process, and has few parameter settings. In detail, the unsupervised feature learning method, sparse autoencoder, is employed to learn relatively small number of convolutional features from input dataset, and then extended features are extracted from the learned features by a multiple normalized difference features extraction method to compose a derivative feature set. At mid-level feature representation stage, in order to avoid poor performance of standard pooling technology in solving problems brought by rotation and translation of scene images, global feature descriptors (histogram moments, mean, variance, standard deviation) are utilized to build mid-level feature representations of images. For validation and comparison purposes, the proposed approach is evaluated via experiments with two challenging high-resolution remote sensing datasets. The results demonstrate that the approach is effective, and shows strong performance for remotely sensed scene classification.

Imitating the behaviors of an arbitrary visual tracking algorithm enables many higher level tasks such as tracker identification and efficient tracker-fusion. It is also useful for discovering the features essential in a black-box tracker or learning from several trackers to form a super-tracker. In this study, we propose a non-linear feature fusion framework, "MIMIC" that imitates many popular trackers by mixing a pool of heterogeneous features. The MIMIC framework consists of two subtasks, feature selection and feature weight tuning. These subtasks, however, tended to suffer from an overfitting problem when the number of videos available for training is limited. To address this issue, we incorporated Dropout algorithm into the training, which grants the trained MIMIC tracker a high degree of generalization. Extensive experiments testified the effectiveness of the proposed framework so that its applications would be promoted into different related tasks in visual tracking.

Image registration is an important preprocessing step in many synthetic aperture radar (SAR) image applications. A key issue in image registration is to reliably establish the correspondences between the feature points extracted from the reference and sensed images. A new point matching algorithm is proposed in this paper to align two SAR images. In the proposed method, by considering image patches as the basic units, a novel local descriptor including the intensity and geometric information is assigned to each feature point, which is more robust to speckle noise. Furthermore, a correspondence establishment scheme is introduced based on the reconstruction errors between feature points calculated by the sparse representation (SR) technique, which is designed for achieving accurate matches. Based on the obtained SR coefficients, a coordinate correction procedure is further proposed for improving the localization accuracy of the obtained correspondences. Both simulated deformed and real SAR images are utilized to evaluate the performance. The experimental results indicate that the proposed method yields a better registration performance in terms of both accuracy and robustness.

Feature selection as one of the key problems of data preprocessing is a hot research topic in pattern recognition, machine learning, and data mining. Evaluating the relevance between features based on information theory is a popular and effective method. However, very little research pays attention to the distinguishing ability of feature, i.e., the degree of a feature distinguishes a given sample with other samples. In this paper, we propose a new feature selection method based on the distinguishing ability of feature. First, we define the concept of maximum-nearest-neighbor, and use this concept to discriminate the nearest neighbors of samples. Then, we present a new measure method for evaluating the quality of feature. Finally, the proposed algorithm is tested on benchmark datasets. Experimental results show that the proposed algorithm can effectively select a discriminative feature subset, and performs as well as or better than other popular feature selection algorithms.

This study proposes an approach that combines edge feature extraction (EFE) with speeded up robust features (SURF) to register rat brain slices. When extracting feature points, using the traditional scale-invariant feature transform (SIFT) approach has several shortcomings, including blurry edge information and the inability to accurately extract smooth edge targets. Therefore, a feature point description method that comprises EFE and SURF is proposed to match local feature points, thereby effectively capturing edge feature points and avoiding mistaking grains and noise in images for feature points. In addition, a wavelet feature description approach is used to resolve incorrect matching caused by noise. Feature point matching for local regions is devised to solve mismatch errors that occur in highly similar regions in an image. Compared to a state-of-the-art SIFT method, the proposed method generates better results for rat brain image registration in terms of subjective visual presentation and registration quality criteria.

Image registration is a preprocessing operation in several synthetic aperture radar (SAR) image applications. View synthesis approach is incorporated in the standard pipeline of feature based image registration, to feed feature detector with additional synthetic views of an image. Incorporating view synthesis improves the registration performance albeit at the cost of additional memory and time to be spent on generation of views and feature extraction across all the views. Hence an iterative approach is proposed which gradually increases the views to be generated till the images are registered. The evaluation results show that the proposed approach achieves high-precision and robust registration of look-angle varied TerraSAR-X images.

Visual object tracking is the process of continuously localizing visual object in a video sequence. We would like to investigate the problem of short-term model-free tracking which the main purpose is to track any object just based on an annotation box of object. Many factors affect the performance of the tracking algorithm. In the Visual Tracker Benchmark, there are eleven challenges in object tracking. There has not been a single tracker that successfully handles all of these scenarios. In addition, the tracker must be fast enough to be useful in real applications. We propose a new tracking algorithm within the Bayesian framework. The proposed algorithm is constructed by solving optimally particle filters (OPF) efficiently using spectral methods. Therefore, the constructed tracker is called as Spectral Tracker (ST). Although ST can efficiently compute object position, it cannot estimate the scale and rotation directly. To overcome this weakness, it is proposed to use multiple observation points simultaneously and to use information on the observation point movement to estimate scale and rotation. In the experiments, the performance of ST tracker was compared with 9 relevant trackers based on 100 data sets. The experimental results on on tracker performance show that increasing performance especially in tracker precision and success rate.

Feature selection has been attracting more attentions in recent years for its advantages in improving the fault diagnosis efficiency and reducing the cost of feature acquisition. In this paper, we regard the feature selection as a clustering process with data decomposition technique and propose a novel feature selection method based on the non-negation matrix factorization (NMF). Alternating Least Squares (ALS) algorithm with sparsity control and decorrelation constrains is adopted to factorize original feature space into two low-rank matrixes (projection vectors and feature spaces). Considering the clustering distribution of the projection space, the optimal feature vectors are calculated by the means of the best updating rule parameters. Besides, the inverse of feature vectors is furtherly utilized in the seeking feature subset, which ensures high classifying performance. Experiments are performed by using two standard data sets and the fault diagnosis of roller bearing case. The results are compared with those obtained by applying the whole feature set and standard feature selection algorithms. The outcomes of comparative analysis have confirmed the effectiveness of the proposed approach.

Purpose The real-time generation of feature descriptors for object recognition is a challenging problem. In this research, the purpose of this paper is to provide a hardware friendly framework to generate sparse features that can be useful for key feature point selection, feature extraction, and descriptor construction. The inspiration is drawn from feature formation processes of the human brain, taking into account the sparse, modular, and hierarchical processing of visual information. Design/methodology/approach A sparse set of neurons referred as active neurons determines the feature points necessary for high-level vision applications such as object recognition. A psycho-physical mechanism of human low-level vision relates edge detection to noticeable local spatial stimuli, representing this set of active neurons. A cognitive memory cell array-based implementation of low-level vision is proposed. Applications of memory cell in edge detection are used for realizing human vision inspired feature selection and leading to feature vector construction for high-level vision applications. Findings True parallel architecture and faster response of cognitive circuits avoid time costly and redundant feature extraction steps. Validation of proposed feature vector toward high-level computer vision applications is demonstrated using standard object recognition databases. The comparison against existing state-of-the-art object recognition features and methods shows an accuracy of 97, 95, 69 percent for Columbia Object Image Library-100, ALOI, and PASCAL VOC 2007 databases indicating an increase from benchmark methods by 5, 3 and 10 percent, respectively. Originality/value A hardware friendly low-level sparse edge feature processing system is proposed for recognizing objects. The edge features are developed based on threshold logic of neurons, and the sparse selection of the features applies a modular and hierarchical processing inspired from the human neural system.

Speed up robust features (SURF) image geometrical registration algorithm available tends to have a one-to-many association problem in feature association. One feature point in an image is associated with multiple feature points in another image, of which some or even all are mismatching points. Coordinates of these mismatching points are used to compute transformation matrix, making it difficult to obtain desirable registration. To solve this problem of SURF algorithm, super-SURF image geometrical registration algorithm is proposed, in which information richness areas are selected to detect feature points and to implement feature points association. The degree of closeness of multiple feature points from the one-to-many feature point pairs is analysed in order to remove feature point pairs with larger errors and retain those with smaller errors. Transformation matrix is then determined with coordinates of feature point pairs retained. Registration image can be obtained after transformation of floating image. Experimental results indicate that super-SURF image geometrical registration algorithm is of higher matching accuracy with less running time than SURF algorithm.

A major challenge in feature matching is the lack of objective criteria to determine corresponding points. Recent methods find match candidates first by exploring the proximity in descriptor space, and then rely on a ratio-test strategy to determine final correspondences. However, these measurements are heuristic and subjectively excludes massive true positive correspondences that should be matched. In this paper, we propose a novel feature matching algorithm for image collections, which is capable of providing quantitative depiction to the plausibility of feature matches. We achieve this by exploring the epipolar consistency between feature points and their potential correspondences, and reformulate feature matching as an optimization problem in which the overall geometric inconsistency across the entire image set ought to be minimized. We derive the solution of the optimization problem in a simple linear iterative manner, where a k-means-type approach is designed to automatically generate consistent feature clusters. Experiments show that our method produces precise correspondences on a variety of image sets and retrieves many matches that are subjectively rejected by recent methods. We also demonstrate the usefulness of the framework in structure from motion task for denser point cloud reconstruction.

The features used in many social media analysis-based applications are usually of very high dimension. Feature selection offers several advantages in highly dimensional cases. Recently, multi-task feature selection has attracted much attention, and has been shown to often outperform the traditional single-task feature selection. Current multi-task feature selection methods are either supervised or unsupervised. In this paper, we address the semi-supervised multi-task feature selection problem. We firstly introduce manifold regularization in multi-task feature selection to utilize the limited number of labeled samples and the relatively large amount of unlabeled samples. However, the graph constructed in manifold regularization from a single feature representation (view) may be unreliable. We thus propose to construct the graph using the heterogeneous feature representations from multiple views. The proposed method is called manifold regularized multi-view feature selection (MRMVFS), which can exploit the label information, label relationship, data distribution, as well as correlation among different kinds of features simultaneously to boost the feature selection performance. All these information are integrated into a unified learning framework to estimate feature selection matrix, as well as the adaptive view weights. Experimental results on three real-world image datasets, NUS-WIDE, Flickr and Animal, demonstrate the effectiveness and superiority of the proposed MRMVFS over other state-of-the-art feature selection methods.

Information fusion is a key step in multimodal biometric systems. The fusion of information can occur at different levels of a recognition system, i.e., at the feature level, matching-score level, or decision level. However, feature level fusion is believed to be more effective owing to the fact that a feature set contains richer information about the input biometric data than the matching score or the output decision of a classifier. The goal of feature fusion for recognition is to combine relevant information from two or more feature vectors into a single one with more discriminative power than any of the input feature vectors. In pattern recognition problems, we are also interested in separating the classes. In this paper, we present discriminant correlation analysis (DCA), a feature level fusion technique that incorporates the class associations into the correlation analysis of the feature sets. DCA performs an effective feature fusion by maximizing the pairwise correlations across the two feature sets and, at the same time, eliminating the between-class correlations and restricting the correlations to be within the classes. Our proposed method can be used in pattern recognition applications for fusing the features extracted from multiple modalities or combining different feature vectors extracted from a single modality. It is noteworthy that DCA is the first technique that considers class structure in feature fusion. Moreover, it has a very low computational complexity and it can be employed in real-time applications. Multiple sets of experiments performed on various biometric databases and using different feature extraction techniques, show the effectiveness of our proposed method, which outperforms other state-of-the-art approaches.

In this paper, we focus on unsupervised feature selection. As we have known, the combination of several feature units into a whole feature vector is broadly adopted for effective object representation, which may inevitably includes some irrelevant/redundant feature units or feature dimensions. Most of the traditional feature selection models can only select the feature dimensions without concerning the intrinsic relationship among different feature units. By taking into consideration the group sparsity of feature dimensions and feature units based on an l 2 , 1 minimization, we propose a new unsupervised feature selection model, unsupervised deep sparse feature selection (UDSFS) in this paper. In comparison with the state-of-the-arts, our UDSFS model can not only select the most discriminative feature units but also assign proper weight to the useful feature dimensions concurrently; moreover, the efficiency and robustness of our UDSFS can be also improved without extracting the discarded irrelevant feature units. For model optimization, we introduce an efficient iterative algorithm to solve the non-smooth, convex model and obtain a global optimization with the convergence rate as O ( 1 / K 2 ) (K is the iteration number). For the experiments, a new medical endoscopic image dataset, Abnormal Endoscopic Image Detection dataset (AEID), is built for evaluation; we also test our model using two public UCI datasets. Various experiments and comparisons with other state-of-the-arts justified the effectiveness and efficiency of our UDSFS model.

Feature-based registration is an effective and most widely used image registration method currently. It includes three critical steps, feature extraction, feature matching and transformation parameters estimation. This paper mainly explores the first two steps. In one of Chahira Serief’s paper about image registration, feature points extraction based on nonsubsampled contourlet transform (NSCT) was proposed and feature points matching based on Zernike moments was adopted. The registration accuracy and robustness of his algorithm are acceptable, but it can still be improved. In this paper, an improved scheme of this registration algorithm is proposed. The rotation invariance of NSCT-based feature points extraction is improved, which is beneficial to extract homologous feature points. And the reliability and effectiveness of Zernike momentsbased feature points matching are improved, which can improve the matching accuracy. The improved registration algorithm can realize registration of images related by larger scaling, rotation and translation transformation. The simulation results show that the registration robustness is further improved, and the registration accuracy is still high.

Local feature extraction methods for images and videos are widely applied in the fields of image understanding and computer vision. However, robust features are detected differently when using the latest feature detectors and descriptors because of diverse image environments. This paper analyzes various feature extraction methods by summarizing algorithms, specifying properties, and comparing performance. We analyze eight feature extraction methods. The performance of feature extraction in various image environments is compared and evaluated. As a result, the feature detectors and descriptors can be used adaptively for image sequences captured under various image environments. Also, the evaluation of feature detectors and descriptors can be applied to driving assistance systems, closed circuit televisions (CCTVs), robot vision, etc.

Automatic image registration is a vital yet challenging task, particularly for multi-sensor remote sensing images. Given the diversity of the data, it is unlikely that a single registration algorithm or a single image feature will work satisfactorily for all applications. Focusing on this issue, the mainly contribution of this paper is to propose an automatic optical-to-SAR image registration method using –level and refinement model: Firstly, a multi-level strategy of coarse-to-fine registration is presented, the visual saliency features is used to acquire coarse registration, and then specific area and line features are used to refine the registration result, after that, sub-pixel matching is applied using KNN Graph. Secondly, an iterative strategy that involves adaptive parameter adjustment for re-extracting and re-matching features is presented. Considering the fact that almost all feature-based registration methods rely on feature extraction results, the iterative strategy improve the robustness of feature matching. And all parameters can be automatically and adaptively adjusted in the iterative procedure. Thirdly, a uniform level set segmentation model for optical and SAR images is presented to segment conjugate features, and Voronoi diagram is introduced into Spectral Point Matching (VSPM) to further enhance the matching accuracy between two sets of matching points. Experimental results show that the proposed method can effectively and robustly generate sufficient, reliable point pairs and provide accurate registration.

A robust algorithm is proposed for tracking a target object in dynamic conditions including motion blurs, illumination changes, pose variations, and occlusions. To cope with these challenging factors, multiple trackers based on different feature representations are integrated within a probabilistic framework. Each view of the proposed multiview (multi-channel) feature learning algorithm is concerned with one particular feature representation of a target object from which a tracker is developed with different levels of reliability. With the multiple trackers, the proposed algorithm exploits tracker interaction and selection for robust tracking performance. In the tracker interaction, a transition probability matrix is used to estimate dependencies between trackers. Multiple trackers communicate with each other by sharing information of sample distributions. The tracker selection process determines the most reliable tracker with the highest probability. To account for object appearance changes, the transition probability matrix and tracker probability are updated in a recursive Bayesian framework by reflecting the tracker reliability measured by a robust tracker likelihood function that learns to account for both transient and stable appearance changes. Experimental results on benchmark datasets demonstrate that the proposed interacting multiview algorithm performs robustly and favorably against state-of-the-art methods in terms of several quantitative metrics.

The key to feature fusion for classification is to take advantage of the discriminative and complementary information from different feature sets, which can be represented as internal (within each feature set) or external structures (across different feature sets). Traditional approaches tend to preserve either internal or external structures via certain feature projection. Some early attempts consider both structures implicitly or indirectly without revealing their relative importance and relevance in feature fusion. We propose a new unsupervised heterogeneous structure fusion (HSF) algorithm that is able to jointly optimize two kinds of structures explicitly and directly via unified feature projection. Specifically, the internal structure is represented based on Locality Preserving Projection (LPP), and the external structure is captured by Canonical Correlation Analysis (CCA). The objective function of HSF combines two feature structures in a closed form which can be optimized alternately via linear programming and eigenvector methods. The HSF solution provides not only the optimal feature projection but also the weights that encode the relative importance between two kinds of feature structures. The experimental results on image classification, face recognition, shape analysis and infrared target recognition demonstrate the effectiveness and efficiency of HSF compared with state-of-the-art methods. Graphical abstractDisplay Omitted HighlightsThe external and internal feature structures are jointly optimized for fusion.Two kinds of feature structures are preserved via a unified feature projection.The nonlinear relationship between two kinds of feature structures is revealed.The new feature fusion algorithm shows great promise in four classification problems.

Feature point matching is a critical step in feature based image registration. For remote sensing images, scale invariant feature transform (SIFT) feature matching is often affected by similar descriptors and there are mismatches. To improve the quality of feature matching and image registration, we propose to use spatial relationship along with the SIFT descriptor for registration. Firstly, initial matches are obtained based on distances between SIFT feature descriptors. Secondly, the spatial relationship of matched points is encoded by kernel partial least squares (KPLS). By analyzing the collinearity of the KPLS features from coarse to fine, false matches are indicated. Finally, correct matches are used to realize accurate registration. Experimental results show an overall significant reduction of the mismatches while maintaining a high rate of correct matches. Compared with several other feature matching methods, the proposed method provides comparable or better results.

An index SSI for measuring the degree of variation of the surface is proposed.A method for identifying potential feature points that are subordinate to different feature lines is present.Existing methods for thinning the potential feature points have been improved. An index of measuring the variation on a surface called the smooth shrink index (SSI) which presents robustness to noise and non-uniform sampling is developed in this work. Afterwards, a new algorithm used for extracting the feature lines was proposed. Firstly, the points with an absolute value of SSI greater than a given threshold are selected as potential feature points. Then, the SSI is applied as the growth condition to conduct region segmentation of the potential feature points. Finally, a bilateral filter algorithm is employed to obtain the final feature points by thinning the potential feature points iteratively. While thinning the potential feature points, the tendency of the feature lines is acquired using principle component analysis (PCA) to restrict the drift direction of the potential feature points, so as to prevent the shrink in the endpoints of the feature lines and breaking of the feature lines induced by non-uniform sampling. Display Omitted

Point clouds collected by terrestrial laser scanning (TLS) from large-scale urban scenes contain a wide variety of objects (buildings, cars, pole-like objects, and others) with symmetric and incomplete structures, and relatively low-textured surfaces, all of which pose great challenges for automatic registration between scans. To address the challenges, this paper proposes a registration method to provide marker-free and multi-view registration based on the semantic feature points extracted. First, the method detects the semantic feature points within a detection scheme, which includes point cloud segmentation, vertical feature lines extraction and semantic information calculation and finally takes the intersections of these lines with the ground as the semantic feature points. Second, the proposed method matches the semantic feature points using geometrical constraints (3-point scheme) as well as semantic information (category and direction), resulting in exhaustive pairwise registration between scans. Finally, the proposed method implements multi-view registration by constructing a minimum spanning tree of the fully connected graph derived from exhaustive pairwise registration. Experiments have demonstrated that the proposed method performs well in various urban environments and indoor scenes with the accuracy at the centimeter level and improves the efficiency, robustness, and accuracy of registration in comparison with the feature plane-based methods.

Image registration deals with establishing correspondences between images of the same scene or object. An image registration algorithm should handle the variations introduced by the imaging system capturing the scene. Scale Invariant Feature Transform (SIFT) is an image registration algorithm based on local features in an image. Compared to the previous registration algorithms, SIFT is more robust to variations caused by changes in size, illumination, rotation, and viewpoint of the images. Owing to its performance, the algorithm is widely studied, modified, and successfully applied in many image and video based applications, in the domains such as medicine, industry, and defense. This paper is an outcome of extensive study on the state-of-art image registration algorithms based on SIFT. Around 20 algorithms based on the SIFT algorithm is discussed. A classification is made based on the objective with which the basic algorithm is modified. A comparative study on the performance, methodology of each...

This letter proposes a simple but effective approach to automatically learn a multilayer image feature for satellite image scene classification. Different from the hand-crafted features which are empirically designed but lack high generalization ability, the proposed approach can autonomously extract the data-dependent feature. The presented feature extraction algorithm is composed of two layers, and the bases of these two layers are uniformly learned by a plain $K$ -means clustering algorithm. Coincidentally, the feature extraction performance of the aforementioned two layers is consistent with visual processing of human visual cortex. More specifically, the first layer can generate edgelike bases, which are analogous to the neuron responses of primary visual cortex (V1), and the second layer can produce cornerlike bases, which resemble the neuron responses of visual extrastriate cortical area two (V2). The proposed feature extraction approach can automatically extract not only simple structure features (e.g., edges) but also complex structure features (e.g., corners and junctions). The learned feature is further discriminated by the linear support vector machine classifier for scene classification. In order to fairly demonstrate the validity of the proposed feature extraction approach, its satellite image scene classification performance is evaluated on the public UCM-21 data set. Experimental results show that the proposed approach can outperform several recent state-of-the-art approaches.

This letter proposes a synthetic aperture radar (SAR) image registration method named Feature-Area Optimization (FAO). First, the traditional area-based optimization model is reconstructed and decomposed into three key but uncertain factors: initialization, slice set and regularization. Next, structural features are extracted by scale invariant feature transform (SIFT) in dual-resolution space (SIFT-DRS), a novel SIFT-Like method dedicated to FAO. Then, the three key factors are determined based on these features. Finally, solving the factor-determined optimization model can get the registration result. A series of experiments demonstrate that the proposed method can register multi-temporal SAR images accurately and efficiently., 5 pages, 5 figures

In this letter, we present an adaptive coarse-to-fine registration method for 3-D indoor mapping using RGB-D data. We weight the 3-D points based on the theoretical random error of depth measurements and introduce a novel disparity-based model for an accurate and robust coarse-to-fine registration. Some feature extraction methods required by the method are also presented. First, our method exploits both visual and depth information to compute the initial transformation parameters. We employ scale-invariant feature transformation for extracting, detecting, and matching 2-D visual features, and their associated depth values are used to perform coarse registration. Then, we use an image-based segmentation technique for detecting regions in the RGB images. Their associated 3-D centroid and the correspondent disparity values are used to refine the initial transformation parameters. Finally, the loop-closure detection and a global adjustment of the complete sequence data are used to recognize when the camera has returned to a previously visited location and minimize the registration errors. The effectiveness of the proposed method is demonstrated with the Kinect data set. The experimental results show that the proposed method can properly map the indoor environment with a relative and absolute accuracy value of around 3-5 cm, respectively.

Non-rigid image registration is an important preprocessing step in synthetic aperture radar (SAR) image applications. A crucial problem that involves in it is to reliably establish the correspondences between the feature points extracted from both the reference and sensed image. In this Letter, a non-rigid registration method is proposed to align two SAR images by registering two sets of feature points extracted from the images. In the proposed method, both point-wise background regional similarity and local spatial constraint are utilised to find correct correspondences between two feature point sets. Point-wise background regional similarity is introduced to enhance feature points similarity measurement. Meanwhile, based on the adjacent spatial relationship between feature point and its neighbouring points, a new concept of local spatial constraint is further proposed to robustly characterise the geometric consistency between them, which is designed to reduce the ambiguous matches aroused by speckle noise and feature outliers. By combining these two improvements, the authors generate a new matching cost function and formulate non-rigid image registration as a correspondence optimisation problem, which can be solved by the probabilistic relaxation method. Experimental results on both simulated and real deformed SAR images indicate the robustness and effectiveness of the proposed method.

For feature extraction, the dimensionality of the feature space is usually much larger than the size of training set. This is known as under sample problem. At this time, local structure is more important than global structure. In this paper, locality sensitive batch feature extraction (LSBFE) is derived based on a new gradient optimization model by exploiting both local and global discriminant structure of data manifold. With the proposed LSBFE, a set of features can be extracted simultaneously. Recognition rate is improved compared with batch feature extraction (BFE), which only considers global information. It is shown that the proposed method achieves good performance for face databases, handwritten digit database, object database and DBWorld data set.

Feature matching, which refers to establishing reliable correspondence between two sets of features (particularly point features), is a critical prerequisite in feature-based registration. In this paper, we propose a flexible and general algorithm, which is called locally linear transforming (LLT), for both rigid and nonrigid feature matching of remote sensing images. We start by creating a set of putative correspondences based on the feature similarity and then focus on removing outliers from the putative set and estimating the transformation as well. We formulate this as a maximum-likelihood estimation of a Bayesian model with hidden/latent variables indicating whether matches in the putative set are outliers or inliers. To ensure the well-posedness of the problem, we develop a local geometrical constraint that can preserve local structures among neighboring feature points, and it is also robust to a large number of outliers. The problem is solved by using the expectation–maximization algorithm (EM), and the closed-form solutions of both rigid and nonrigid transformations are derived in the maximization step. In the nonrigid case, we model the transformation between images in a reproducing kernel Hilbert space (RKHS), and a sparse approximation is applied to the transformation that reduces the method computation complexity to linearithmic. Extensive experiments on real remote sensing images demonstrate accurate results of LLT, which outperforms current state-of-the-art methods, particularly in the case of severe outliers (even up to 80%).