Local Feature Vectors Research Articles

Saliency-based Pedestrian Detection in Far Infrared Images

Night time pedestrian detection is more and more important in advanced driver assistant systems (ADAS). Traditional pedestrian detection algorithms in far infrared (FIR) images lack accuracy and have long processing times. Focusing on this issue, in this paper, a visual saliency-based pedestrian detection algorithm is proposed. First, areas that contain suspected pedestrians are detected using a fusion saliency-based method. Then, the sub-image of the suspected pedestrian is used as an input to a histogram of local intensity difference feature and cross kernel-based support vector machine classifier to make a final determination. Experiments performed using a real FIR road image data set demonstrated that the proposed fusion saliency-based region of interest (ROI) detection method has the largest pedestrian inclusion rate and the smallest ROI proportion compared with three other methods. Besides, compared with existing state-of-the-art pedestrian detection algorithms, the proposed method demonstrates a much higher pedestrian detection rate with a comparably short processing time.

Fast Registration Methodology for Fastener Assembly of Large-Scale Structure

Fastener assembly is a tedious and time-consuming work because operators have to check assembly manuals and find right fastener for each hole. Hence, this paper aims to develop a three-dimensional (3-D) projection system that projects assembly instruction onto the work piece surface directly to guide operators to assemble. However, in order to project the instruction accurately, the corresponding part of the computer-aided design model of the physical scanned area needs to be attained through the rapid and accurate registration. In order to achieve this goal, first, a high-accuracy and rapid 3-D measurement system is developed; second, a fast registration method based on local multiscale geometric feature vector is proposed to accelerate the registration speed and improve the registration reliability. Experimental results demonstrate the measurement accuracy of the developed system, and verify the feasibility of the proposed registration method. Hence, the proposed method can lead to improved assembly efficiency and decreased error probability, making great contributions to large-scale structure assembly.

A Novel Texture Feature Description Method Based on the Generalized Gabor Direction Pattern and Weighted Discrepancy Measurement Model

Texture feature description is a remarkable challenge in the fields of computer vision and pattern recognition. Since the traditional texture feature description method, the local binary pattern (LBP), is unable to acquire more detailed direction information and always sensitive to noise, we propose a novel method based on generalized Gabor direction pattern (GGDP) and weighted discrepancy measurement model (WDMM) to overcome those defects. Firstly, a novel patch-structure direction pattern (PDP) is proposed, which can extract rich feature information and be insensitive to noise. Then, motivated by searching for a description method that can explore richer and more discriminant texture features and reducing the local Gabor feature vector’s high dimension problem, we extend PDP to form the GGDP method with multi-channel Gabor space. Furthermore, WDMM, which can effectively measure the feature distance between two images, is presented for the classification and recognition of image samples. Simulated experiments on olivetti research laboratory (ORL), Carnegie Mellon University pose, illumination, and expression (CMUPIE) and Yale B face databases under different illumination or facial expression conditions indicate that the proposed method outperforms other existing classical methods.

A method of 3D CAD model retrieval based on spatial bag of words

In order to improve both the discriminative power for models' local parts and the searching efficiency in 3D CAD model retrieval, a novel hierarchical feature descriptor for retrieval based on spatial bag of words is proposed in this paper. By extracting the essential information from Boundary Representation (B-Rep), 3D CAD models are transformed to Labelled Attribute Adjacency Graphs (LAAGs). Next, the models in training dataset are segmented into different regions according to their corresponding LAAG with an improved segmentation method. All collections of these local regions are described as local feature vectors with graph spectrum, and the codebook is created by clustering all these vectors. Each library model is then decomposed with the same methods mentioned above and globally represented as a spatial histogram of word pairs along with the adjacent relations of its regions, called Spatial Bags-of-Words (SBoWs),and then, the hierarchical feature descriptor(HFD) of each library model composed of global SBoWs and local graph spectrum is constructed. Finally, according to HFD, a two-level searching framework is presented for CAD model retrieval: the candidates are acquired by comparing the query with each target model based on their SBoWs vectors, and the remaining candidates are verified using optimal matching algorithm according to their local features. Experimental results show that the proposed methods promote both retrieval quality and efficiency significantly, so they can support the effective reuse of CAD models for enterprises.

Combining of Local Histogram and Local Slice Feature Vectors for Classifying Diffuse Lung Opacities in Thin-Section Computed Tomography Images

Combining of Local Histogram and Local Slice Feature Vectors for Classifying Diffuse Lung Opacities in Thin-Section Computed Tomography Images

Copy–Move Forgery Detection by Matching Triangles of Keypoints

Copy–move forgery is one of the most common types of tampering for digital images. Detection methods generally use block-matching approaches, which first divide the image into overlapping blocks and then extract and compare features to find similar ones, or point-based approaches, in which relevant keypoints are extracted and matched to each other to find similar areas. In this paper, we present a very novel hybrid approach, which compares triangles rather than blocks, or single points. Interest points are extracted from the image, and objects are modeled as a set of connected triangles built onto these points. Triangles are matched according to their shapes (inner angles), their content (color information), and the local feature vectors extracted onto the vertices of the triangles. Our methods are designed to be robust to geometric transformations. Results are compared with a state-of-the-art block matching method and a point-based method. Furthermore, our data set is available for use by academic researchers.

Discriminative quadratic feature learning for handwritten Chinese character recognition

In this paper, we propose a feature learning method for handwritten Chinese character recognition (HCCR), called discriminative quadratic feature learning (DQFL). Based on original gradient direction feature representation, quadratic correlation between features is used to promote the feature dimensionality, then discriminative feature extraction (DFE) is used for dimensionality reduction. By combining dimensionality promotion and reduction, we can learn a much more discriminative and nonlinear feature representation, which can then boost the classification accuracy significantly. For dimensionality promotion, two types of correlation are exploited, namely, statistical correlation and spatial correlation. Statistical correlation is computed on multiple local feature vectors in different regions of the character image; while spatial correlation encodes the dependency between features of two positions. Feature correlation increases the dimensionality by over 40,000. DFE then reduces the dimensionality to less than 300 without losing discriminability. Classification is performed using nearest prototype classifier (NPC), modified quadratic discriminant function (MQDF) and discriminative learning quadratic discriminant function (DLQDF). In experiments on the CASIA-HWDB1.1 standard dataset, the proposed DQFL method improves the test accuracies of NPC, MQDF and DLQDF by 4.94%, 1.83%, and 1.82%, respectively. The test accuracy is further improved by training set expansion. On the ICDAR 2013 Chinese handwriting recognition competition dataset, the proposed DQFL+DLQDF classifier outperforms the best participating system based on deep convolutional neural network (CNN), while the test speed is much faster.

A novel texture description by local multi-channels Gabor comprehensive patterns

The texture description is an important research topic in the pattern recognition region. A new texture description algorithm using Local Multi-channels Gabor Comprehensive Patterns (LMGCP) was proposed. The motivation for the LMGCP model is to find more rich and discriminant texture measurement and deal with the high dimension problem of the local Gabor feature vector. First, the sample image was filtered by multi-orientation Gabor filters with multi-scale to extract their corresponding local Gabor magnitude map (LGMM). Second, the Local Comprehensive Patterns (LCP) were presented, which can compute the relationship between the referenced pixel and its neighbors by encoding gray-level difference based on 0°, 45°, 90°, 135°, 180°, 225°, 270°, 315° high orders direction derivatives patterns (DDP) and the direction magnitude patterns (DMP). Finally, the dimension of local Gabor feature vector is higher, so the LGMM were transformed by LCP named by LMGCP, which were concatenated into and feature vector. Simulated experiments and comparisons on subsets of Yale B and CMUPIE face databases under ideal condition, different illumination condition, different facial expression and partial occlusion show that the proposed algorithm is an outstanding method better than the LBP, the local derivative patterns, and the LTP, Local Gabor Transform.

3D Ear Shape Matching Using Joint α-Entropy

In this article, we investigate the use of joint α-entropy for 3D ear matching by incorporating the local shape feature of 3D ears into the joint α-entropy. First, we extract a sufficient number of key points from the 3D ear point cloud, and fit the neighborhood of each key point to a single-value quadric surface on product parameter regions. Second, we define the local shape feature vector of each key point as the sampling depth set on the parametric node of the quadric surface. Third, for every pair of gallery ear and probe ear, we construct the minimum spanning tree (MST) on their matched key points. Finally, we minimize the total edge weight of MST to estimate its joint α-entropy — the smaller the entropy is, the more similar the ear pair is. We present several examples to demonstrate the advantages of our algorithm, including low time complexity, high recognition rate, and high robustness. To the best of our knowledge, it is the first time that, in computer graphics, the classical information theory of joint α-entropy is used to deal with 3D ear shape recognition.

Automatic Aerial Target Recognition using A Robust SURF-MSER Feature-based Algorithm

Recognition and Identification of Targets in Aerial images play a significant role and wide applications in imaging guidance of cruise missiles, standoff missiles and during terminal phase of ballistic missiles from optics images or radar images. A proposed scene matching algorithm based on both feature detection and extraction Speeded-Up Robust Features (SURF) and Maximally Stable Extremal Regions (MSER) techniques. The extracted features are invariant to image scaling, translation, rotation, and partially invariant to illumination changes and affine or 3D projection. Image data set references of target are collected based on the criteria of azimuth, elevation and altitude of the target image region of interest. The proposed targets are assumed to be buildings. Firstly, SURF and MSER local feature vectors are detected and extracted from reference images of the target and merged to constitute the feature vectors matrix offline. Secondly, a feature detection and extraction using both SURF and MSER algorithms for the real time captured image are performed to form real time feature matrix. Finally, a matching algorithm is applied between both matrices of reference images and real time image. Moreover, Key points outlier elimination technique is used based on a Random Sample Consensus (RANSAC) algorithm to reduce false matching and speed up the matching procedure. The proposed algorithm has been tested on different real image data sets. A comparative analysis is performed between the SURF, FAST corner detector and MSER individually with the proposed SURF-MSER algorithm. The proposed algorithm process more complicated deformation between reference image and real time acquired image and attained higher matching accuracy.

Application of LCD-SVD Technique and CRO-SVM Method to Fault Diagnosis for Roller Bearing

Targeting the nonlinear and nonstationary characteristics of vibration signal from fault roller bearing and scarcity of fault samples, a novel method is presented and applied to roller bearing fault diagnosis in this paper. Firstly, the nonlinear and nonstationary vibration signal produced by local faults of roller bearing is decomposed into intrinsic scale components (ISCs) by using local characteristic-scale decomposition (LCD) method and initial feature vector matrices are obtained. Secondly, fault feature values are extracted by singular value decomposition (SVD) techniques to obtain singular values, while avoiding the selection of reconstruction parameters. Thirdly, a support vector machine (SVM) classifier based on Chemical Reaction Optimization (CRO) algorithm, called CRO-SVM method, is designed for classification of fault location. Lastly, the proposed method is validated by two experimental datasets. Experimental results show that the proposed method based LCD-SVD technique and CRO-SVM method have higher classification accuracy and shorter cost time than the comparative methods.

Human Action Recognition Based on Fusion Features Extraction of Adaptive Background Subtraction and Optical Flow Model

A novel method based on hybrid feature is proposed for human action recognition in video image sequences, which includes two stages of feature extraction and action recognition. Firstly, we use adaptive background subtraction algorithm to extract global silhouette feature and optical flow model to extract local optical flow feature. Then we combine global silhouette feature vector and local optical flow feature vector to form a hybrid feature vector. Secondly, in order to improve the recognition accuracy, we use an optimized Multiple Instance Learning algorithm to recognize human actions, in which an Iterative Querying Heuristic (IQH) optimization algorithm is used to train the Multiple Instance Learning model. We demonstrate that our hybrid feature-based action representation can effectively classify novel actions on two different data sets. Experiments show that our results are comparable to, and significantly better than, the results of two state-of-the-art approaches on these data sets, which meets the requirements of stable, reliable, high precision, and anti-interference ability and so forth.

GMM-based intermediate matching kernel for classification of varying length patterns of long duration speech using support vector machines.

Dynamic kernel (DK)-based support vector machines are used for the classification of varying length patterns. This paper explores the use of intermediate matching kernel (IMK) as a DK for classification of varying length patterns of long duration speech represented as sets of feature vectors. The main issue in construction of IMK is the choice for the set of virtual feature vectors used to select the local feature vectors for matching. This paper proposes to use components of class-independent Gaussian mixture model (CIGMM) as a representation for the set of virtual feature vectors. For every component of CIGMM, a local feature vector each from the two sets of local feature vectors that has the highest probability of belonging to that component is selected and a base kernel is computed between the selected local feature vectors. The IMK is computed as the sum of all the base kernels corresponding to different components of CIGMM. It is proposed to use the responsibility term weighted base kernels in computation of IMK to improve its discrimination ability. This paper also proposes the posterior probability weighted DKs (including the proposed IMKs) to improve their classification performance and reduce the number of support vectors. The performance of the support vector machine (SVM)-based classifiers using the proposed IMKs is studied for speech emotion recognition and speaker identification tasks and compared with that of the SVM-based classifiers using the state-of-the-art DKs.

Contextual Patch Feature Learning for Face Recognition

Local features, such as local binary patterns(LBP), have shown better performance than global featurein the problem of face recognition. However, the methodsto extract the local features are usually given as fixed,and also neglect the class labels of the training samples.In this paper, we propose a novel algorithm to learn adiscriminate local feature from the small patches of theface image to boost the face recognition. The pixels of eachimage patch and its neighboring patches are both used toconstruct the local feature. The pixel vector of each patchis mapped to new subspaces by a transformation matrix,and mapped pixel vectors the neighboring patches are alsocombined to obtain the local feature vector. The subspacemapping parameter and the neighboring patch combinationparameter are learned to minimize the distances of localfeatures between the same person, and at the same timeto maximize that between different persons. We performexperiments on some benchmark face image database toshow the advantage of the proposed method.

Palmprint Recognition Based on Local Fisher Discriminant Analysis

A new palmprint recognition method based on local Fisher discriminant analysis(LFDA) is proposed. In order to solve the singularity of the eigenvalue equation matrix in small-size-sample cases such as image recognition, image down-sample is first used to reduce the palmprint space dimensionality. The LFDA is applied to extract the low projection vectors. Then the training images and test images are projected onto the projection vectors to get the local palmprint feature vectors. Finally, the cosine distance between two feature vectors is calculated to match palmprint. The new algorithm is tested in PolyU plmprint database. The results show that compared with principal component analysis (PCA), Fisher discriminant analysis (FDA), independent component analysis (ICA), and kernel principal component analysis (KPCA), the recognition rate of the new algorithm is the highest which is 98.95%, and the recognition time is 0.031s, so it meets the real-time system specification.

HMM Based Intermediate Matching Kernel for Classification of Sequential Patterns of Speech Using Support Vector Machines

In this paper, we address the issues in the design of an intermediate matching kernel (IMK) for classification of sequential patterns using support vector machine (SVM) based classifier for tasks such as speech recognition. Specifically, we address the issues in constructing a kernel for matching sequences of feature vectors extracted from the speech signal data of utterances. The codebook based IMK and Gaussian mixture model (GMM) based IMK have been proposed earlier for matching the varying length patterns represented as sets of features vectors for tasks such as image classification and speaker recognition. These methods consider the centers of clusters and the components of GMM as the virtual feature vectors used in the design of IMK. As these methods do not use sequence information in matching the patterns, these methods are not suitable for matching sequential patterns. We propose the hidden Markov model (HMM) based IMK for matching sequential patterns of varying length. We consider two approaches to design the HMM-based IMK. In the first approach, each of the two sequences to be matched is segmented into subsequences with each subsequence aligned to a state of the HMM. Then the HMM-based IMK is constructed as a combination of state-specific GMM-based IMKs that match the subsequences aligned with the particular states of the HMM. In the second approach, the HMM-based IMK is constructed without segmenting sequences, and by matching the local feature vectors selected using the responsibility terms that account for being in a state and generating the feature vectors by a component of the GMM of that state. We study the performance of the SVM based classifiers using the proposed HMM-based IMK for recognition of isolated utterances of E-set in English alphabet and recognition of consonent–vowel segments in Hindi language.

Integration of Global and Local Salient Features for Scene Modeling in Mobile Robot Applications

Many approaches have recently used global image descriptors and/or local key-point descriptors for scene understanding. In fact these approaches have suffered from lack of spatial information by using local key-point descriptors, and lack of viewpoint and local information by using global image descriptors. To overcome these problems, this paper addresses a novel image descriptor based on salient line segments (SLS), in which the global and local image features are integrated into low dimensional feature vectors. In this descriptor, low level feature maps are first computed in four scales by applying a center-surround competition technique to enhance the dominant edges and suppress small line segments. These maps are then used to extract the SLS of the image patches by creating histogram of gradients in the receptive cells. Afterwards, the global features are formed into a single vector from the coarser scales of the SLSs, and the local feature vectors are formed from the frequency of the appearance of SLSs in the finer scale. Finally, a classification step recognizes the scene of an input image by applying multi-class SVM with a Radial Bias Function (RBF) kernel. The system is performed on image sequences taken from natural scenes by a mobile agent under controlled and unexpected changes in environmental conditions. Experiments on image datasets show that the proposed method is able to classify the scenes more accurately than former methods in mobile agent environments.

Class-specific GMM based intermediate matching kernel for classification of varying length patterns of long duration speech using support vector machines

Dynamic kernel based support vector machines are used for classification of varying length patterns. This paper explores the use of intermediate matching kernel (IMK) as a dynamic kernel for classification of varying length patterns of long duration speech represented as sets of feature vectors. The main issue in construction of IMK is the choice for the set of virtual feature vectors used to select the local feature vectors for matching. The components of class-independent GMM (CIGMM) have been used earlier as a representation for the set of virtual feature vectors. For every component of CIGMM, a local feature vector each from the two sets of local feature vectors that has the highest probability of belonging to that component is selected and a base kernel is computed between the selected local feature vectors. The IMK is computed as the sum of all the base kernels corresponding to different components of CIGMM. The construction of CIGMM-based IMK does not use the class-specific information, as the local feature vectors are selected using the components of CIGMM that is common for all the classes. We propose two novel methods to build a better discriminatory IMK-based SVM classifier by considering a set of virtual feature vectors specific to each class depending on the approaches to multiclass classification using SVMs. In the first method, we propose a class-wise IMK based SVM for every class by using components of GMM built for a class as the set of virtual feature vectors for that class in the one-against-the-rest approach to multiclass pattern classification. In the second method, we propose a pairwise IMK based SVM for every pair of classes by using components of GMM built for a pair of classes as the set of virtual feature vectors for that pair of classes in the one-against-one approach to multiclass classification. We also proposed to use the mixture coefficient weighted and responsibility term weighted base kernels in computation of class-specific IMKs to improve their discrimination ability. This paper also proposes the posterior probability weighted dynamic kernels to improve their classification performance and reduce the number of support vectors. The performance of the SVM-based classifiers using the proposed class-specific IMKs is studied for speech emotion recognition and speaker identification tasks and compared with that of the SVM-based classifiers using the state-of-the-art dynamic kernels.

Performance Improvement of Face Recognition System using Selective Local Feature Vectors

Face recognition is one of the important applications of image processing and it has gained significant attention in wide range of law enforcement areas in which security is of prime concern. Although the existing automated machine recognition systems have certain level of maturity but their accomplishments are limited due to real time challenges. Face recognition systems are impressively sensitive to appearance variations due to lighting, expression and aging. The major metric in modeling the performance of a face recognition system is its accuracy of recognition. This paper proposes a novel method which improves the recognition accuracy as well as avoids face datasets being tampered through image splicing techniques. Proposed method uses a non-statistical procedure which avoids training step for face samples thereby avoiding generalizability problem which is caused due to statistical learning procedure. This proposed method performs well with images with partial occlusion and images with lighting variations as the local patch of the face is divided into several different patches. The performance improvement is shown considerably high in terms of recognition rate and storage space by storing train images in compressed domain and selecting significant features from superset of feature vectors for actual recognition.

Probabilistic neural network with homogeneity testing in recognition of discrete patterns set

The article is devoted to pattern recognition task with the database containing small number of samples per class. By mapping of local continuous feature vectors to a discrete range, this problem is reduced to statistical classification of a set of discrete finite patterns. It is demonstrated that the Bayesian decision under the assumption that probability distributions can be estimated using the Parzen kernel and the Gaussian window with a fixed variance for all the classes, implemented in the PNN, is not optimal in the classification of a set of patterns. We presented here the novel modification of the PNN with homogeneity testing which gives an optimal solution of the latter task under the same assumption about probability densities. By exploiting the discrete nature of patterns our modification prevents the well-known drawbacks of the memory-based approach implemented in both the PNN and the PNN with homogeneity testing, namely, low classification speed and high requirements to the memory usage. Our modification only requires the storage and processing of the histograms of input and training samples. We present the results of an experimental study in two practically important tasks: (1) the problem of Russian text authorship attribution with character n-grams features; and (2) face recognition with well-known datasets (AT&T, FERET and JAFFE) and comparison of color- and gradient-orientation histograms. Our results support the statement that the proposed network provides better accuracy (1%–7%) and is much more resistant to change of the smoothing parameter of Gaussian kernel function in comparison with the original PNN.