Haihong Hu

Frame difference energy image for gait recognition with incomplete silhouettes

The quality of human silhouettes has a direct effect on gait recognition performance. This paper proposes a robust dynamic gait representation scheme, frame difference energy image (FDEI), to suppress the influence of silhouette incompleteness. A gait cycle is first divided into clusters. The average image of each cluster is denoised and becomes the dominant energy image (DEI). FDEI representation of a frame is constructed by adding the corresponding clusters DEI and the positive portion of the frame difference between the former frame and the current frame. FDEI representation can preserve the kinetic and static information of each frame, even when the silhouettes are incomplete. This proposed representation scheme is tested on the CMU Mobo gait database with synthesized occlusions and the CASIA gait database (dataset B). The frieze and wavelet features are adopted and hidden Markov model (HMM) is employed for recognition. Experimental results show the superiority of FDEI representation over binary silhouettes and some other algorithms when occlusion or body portion lost appears in the gait sequences.

Factorial HMM and Parallel HMM for Gait Recognition

Information fusion offers a promising solution to the development of a high-performance classification system. In this paper, the problem of multiple gait features fusion is explored with the framework of the factorial hidden Markov model (FHMM). The FHMM has a multiple-layer structure and provides an alternative to combine several gait features without concatenating them into a single augmented feature. Besides, the feature concatenation is used to directly concatenate the features and the parallel HMM (PHMM) is introduced as a decision-level fusion scheme, which employs traditional fusion rules to combine the recognition results at decision level. To evaluate the recognition performances, McNemars test is employed to compare the FHMM feature-level fusion scheme with the feature concatenation and the PHMM decision-level fusion scheme. Statistical numerical experiments are carried out on the Carnegie Mellon University motion of body and the Institute of Automation of the Chinese Academy of Sciences gait databases. The experimental results demonstrate that the FHMM feature-level fusion scheme and the PHMM decision-level fusion scheme outperform feature concatenation. The FHMM feature-level fusion scheme tends to perform better than the PHMM decision-level fusion scheme when only a few gait cycles are available for recognition.

Gait recognition using hidden markov model

Gait-based human identification is a challenging problem and has gained significant attention. In this paper, a new gait recognition algorithm using Hidden Markov Model (HMM) is proposed. The input binary silhouette images are preprocessed by morphological operations to fill the holes and remove noise regions. The width vector of the outer contour is used as the image feature. A set of initial exemplars is constructed from the feature vectors of a gait cycle. The similarity between the feature vector and the exemplar is measured by the inner product distance. A HMM is trained iteratively using Viterbi algorithm and Baum-Welch algorithm and then used for recognition. The proposed method reduces image feature from the two-dimensional space to a one-dimensional vector in order to best fit the characteristics of one-dimensional HMM. The statistical nature of the HMM makes it robust to gait representation and recognition. The performance of the proposed HMM-based method is evaluated using the CMU MoBo database.

Clustering Validity Based on the Improved Hubert \Gamma Statistic and the Separation of Clusters

The validity of clustering is one important research field in clustering analysis, and many clustering validity functions have been proposed, especially those based on the geometrical structure of data set, such as Dunns index and Xie-Beni index. In this way, the compactness and the separation of clusters are usually taken into account. Xie-Beni index decreases with the number of partitions increasing. It is difficult to choose the optimal number of clusters when there are lots of clusters in data. In this paper, a novel clustering validity function is proposed, which is based on the improved Huber Gamma statistic combined with the separation of clusters. Unlike other clustering validity, the function has the only maximum with the clustering number increasing. The experiments indicate that the function can be used as the optimal index for the choice of the clustering numbers

Factorial hidden Markov models for gait recognition

Gait recognition is an effective approach for human identification at a distance. During the last decade, the theory of hidden Markov models (HMMs) has been used successfully in the field of gait recognition. However the potentials of some new HMM extensions still need to be exploited. In this paper, a novel alternative gait modeling approach based on Factorial Hidden Markov Models (FHMMs) is proposed. FHMMs are of a multiple layer structure and provide an interesting alternative to combining several features without the need of collapse them into a single augmented feature. We extracted irrelated features for different layers and iteratively trained its parameters through the Expectation Maximization (EM) algorithm and Viterbi algorithm. The exact Forward-Backward algorithm is used in the E-step of EM algorithm. The performances of the proposed FHMM-based gait recognition method are evaluated using the CMU MoBo database and compared with that of HMMs based methods.

Appearance-Based gait recognition using independent component analysis

For human identification at distance (HID) applications, gait characteristics are hard to conceal and has the inherent merits such as non-contact and unobtrusive. In this paper, a novel appearance-based method for automatic gait recognition is proposed using independent component analysis (ICA). Principal component analysis (PCA) is performed on image sequences of all persons to get the uncorrelated PC coefficients. Then, ICA is performed on the PC coefficients to obtain the more independent IC coefficients. The IC coefficients from the same person are averaged and the mean coefficients are used to represent individual gait characteristics. For improving computational efficiency, a fast and robust method named InfoMax algorithm is used for calculating independent components. Gait recognition performance of the proposed method was evaluated by using CMU MoBo dataset and USF Challenge gait dataset. Experiment results show the efficiency and advantages of the proposed method.

Human Gait Recognition Based on Kernel Independent Component Analysis

In this paper we present a feature representation method based on Kernel Independent Component Analysis for gait recognition. The Kernel ICA combines the strengths of both Kernel and Independent Component Analysis (ICA) approaches. Principal Component Analysis (PCA) is performed as a preparation for Kernel ICA, and then we use Kernel ICA algorithm to obtain the Independent Components (IC). The mean IC coefficients are used to represent different gaits. We compare the performance of Kernel ICA with some classical algorithms such as FastICA and Baseline etc. within the context of appearance- based gait recognition problem using the CMU MoBo database and USF Challenge database. Experimental results show that Kernel ICA based method gives a competitive performance in both accuracy and convergence speed in gait recognition problem.

Multi-atlas registration and adaptive hexahedral voxel discretization for fast bioluminescence tomography

Bioluminescence tomography (BLT) has been a valuable optical molecular imaging technique to non-invasively depict the cellular and molecular processes in living animals with high sensitivity and specificity. Due to the inherent ill-posedness of BLT, a priori information of anatomical structure is usually incorporated into the reconstruction. The structural information is usually provided by computed tomography (CT) or magnetic resonance imaging (MRI). In order to obtain better quantitative results, BLT reconstruction with heterogeneous tissues needs to segment the internal organs and discretize them into meshes with the finite element method (FEM). It is time-consuming and difficult to handle the segmentation and discretization problems. In this paper, we present a fast reconstruction method for BLT based on multi-atlas registration and adaptive voxel discretization to relieve the complicated data processing procedure involved in the hybrid BLT/CT system. A multi-atlas registration method is first adopted to estimate the internal organ distribution of the imaged animal. Then, the animal volume is adaptively discretized into hexahedral voxels, which are fed into FEM for the following BLT reconstruction. The proposed method is validated in both numerical simulation and an in vivo study. The results demonstrate that the proposed method can reconstruct the bioluminescence source efficiently with satisfactory accuracy.

Lateral inhibition network model optimization by evolutionary strategy for image segmentation

Image segmentation is a fundamental image processing technology. There are many kinds of image segmentation methods, but most of them are problem oriented. In this paper, image segmentation method based on lateral inhibition network is presented. Lateral inhibition network is a biological vision model. When an image is filtered by a lateral inhibition network, its low frequency components are inhibited while the high frequency components are enhanced. The lateral inhibited image is much easier to be segmented because of its increased inter-class difference and decreased intra-class difference. The parameters of the lateral inhibition network model determine the inhibited image, thus affect the image segmentation result greatly. But there are no assured rules to determine the parameters. We propose an evolutionary strategy (ES) based method to search the optimal weighting parameters of the lateral inhibition network model. The objective function of ES is a multiattribute fitness function that combines multiple criteria of clustering and entropy information. The original image is filtered using the optimal lateral inhibition network and then the inhibited image is segmented by an optimized threshold. Using test images of various characteristics, the proposed method is evaluated by four objective image segmentation evaluation indexes. The experimental results show its validity and universality.