Xinman Zhang

HALCON application for shape-based matching

Shape-based matching is mainly emphasized in this paper for actual workpiece applications based HALCON, a new machine vision software. First a measuring ROI, i.e., the region of interest is determined to create a model. Then the matching and locating of the search object is achieved by using affine transform. Experiment result indicates that this method can find objects starting based on a single model image and localize objects with subpixel accuracy.

Multifocus image restoration and fusion method based on genetic search strategies

An adaptive genetic search algorithm was developed which applied to fused image blocks for clear restoration of two spatially registered multifocus images from same scene. In this method, the size of image block is defined as chromosome, after crossover and mutation, the global optimal solution would be got. Three kinds of quantitative evaluation criteria are used on the analysis and effect evaluation of different fused images, such as root mean square error, entropy and mutual information. Two cases are discussed and experiments demonstrate that in one case most fused images can achieve perfect reconstruction to the reference image when the focus objects are not overlapped blurred, and in another case this method performs better when the focus objects are overlapped blurred than that of the other two methods.

Dempster-Shafer reasoning with application multisensor object recognition system

Firstly by analyzing comprehensively the different sensor acquisitions based on normal distribution, a mathematical model is established to attain the basic probability assignments. Following this a multisensor data fusion method-based Dempster-Shafer reasoning is proposed to resolve object recognition problems. Offered by multilevel accumulation of assignments recursively the fusion estimates based on global information is obtained to testify the recognition performance. It is optimal than that of a single sensor with an average decline 74 percent of uncertainty value in a case study of pyrites recognition system, thereby demonstrating the effectiveness and correctness of this approach.

Ridgelet and BP neural network based face detection method

The ridgelet transformation was introduced as a sparse expansion for functions on continuous spaces that are smooth away from discontinuities along lines. Focusing on the face detection problem, a novel face detection method using ridgelet transform and BP neural network was proposed. Two stages are involved in this method. Firstly, the pretreated face images are decomposed by the ridgelet transformation. Then the corresponding ridgelet coefficients are set as the input samples for a well-designed BP neural network. The good results are thus obtained.

A new method of medical image fusion based on nonsubsampled contourlet transform

To improve the normal medical image fusion algorithm in order to avoid the loss of the detailed information in the processes of medical image fusion, a multiscale medical image fusion method based on nonsubsampled contourlet transform(NSCT) is proposed in this paper. First, the source images(MRI and CT images) are decomposed by using nonsubsampled contourlet transform. Then, the details of contourlet coefficients are fused on each corresponding levels with a vision feature fusion operator. Finally, the fused image will be obtained by taking the inverse nonsubsampled contourlet transformation. The experimental results show that the effect of the nonsubsampled contourlet-based method is obviously improved, and the proposed method can effectively preserve the detailed information of the source images.

The Finite Ridgelet Image Fusion Scheme by Combining Remote Sensing Images

The ridgelet transform was introduced as a sparse expansion for functions on continuous spaces that were smooth away from discontinuities along lines. Based on the idea of ridgelet a new fusion method combined dyadic wavelet transform was proposed. Numerical results show that the FRIT is more effective than the wavelet transform in image fusion by combing remote sensing images

Biometric recognition using digital curvelet transform and BP neural network

The theoretical studies indicate digital curvelet transform to be an even better method than wavelets for optical application. In this paper, a multiscale biometric recognition method based on digital curvelet transform via wrapping is surveyed and studied. First, all images are decomposed by using curvelet transform. As a result of performing curvelet transform, curvelet coefficients of low frequency and high frequency in different scales and various angels will be obtained. Then, low frequency coefficients as study samples to the BP neural network are applied. Finally, low frequency coefficients of testing image are used to simulate neural network, then recognition results will be obtained. The experiments are performed on the Cambridge University ORL database, and the results show that the recognition rate of the curvelet-based method is obviously improved.

3D Measurement of Specular Reflection Surface by Learning SFS Algorithm-Based RBF Model

Directing to specular reflection surface a SFS algorithm-based RBF specular model is surveyed and studied to reconstruct a profile. In this paper detailed algorithm and discrete equations are discussed and practical application results on synthetic images demonstrate the algorithm performance, ever when the lighting conditions or camera parameters are uncertain.