Font Size: a A A

Research On The Methods And Application Of Machine Visison Inspection With Texture Analysis

Posted on:2011-02-22Degree:DoctorType:Dissertation
Country:ChinaCandidate:H X SunFull Text:PDF
GTID:1118360308485578Subject:Instrument Science and Technology
Abstract/Summary:PDF Full Text Request
Machine vision is a technology of simulating human vision to detect, measure and control objects by using computer image processing systems. It has great advantage and development prospect, and has been widely used in the fields of industry, agriculture, military affairs, and national defense and so on. Vision inspection is an application in detection area by means of machine vision theory and technique. Moreover, texture defect detection is one of important and difficult problems in vision inspection. Several key techniques about texture feature extraction method, texture defect detection method and their applications in real vision inspection systems are studied in this paper. The brief structure of the research and the novel approaches are as follows:The sequency characteristics of sequence are used to describe the image texture. A new multiresolution and rotation invariant texture descriptor is proposed based on the Local Walsh Spectrum (LWS). In vision inspection, the results are frequently affected by interference of image rotation. Moreover, one-scale texture descriptors can not represent the surface characteristic of detecting targets. The rotation invariant of the proposed texture descriptors can be achieved because of circular-shift-invariant of the discrete Walsh transform power spectrum. Simultaneously, the multiresolution texture features can be obtained by LWS. Furthermore, based on the sequency characteristic, the two-family sequency LWS (TSLWS) descriptor is proposed, and the relationship between LWS and Local Binary Pattern (LBP) is revealed. The results of texture classification and segmentation experiments show that LWS and TSLWS have the satisfying texture discrimination performance and rotation invariant.The utilization of the phase information of complex wavelet coefficient is studied in image texture feature extraction firstly. A novel method of texture description is presented combining the coefficient phase and amplitude of Dual-Tree Complex Wavelet Transform (DT-CWT). With overcame the shortcomings of classical real Discrete Wavelet Transform (DWT), DT-CWT can express the phase information of decomposition coefficient. But it is difficulty that how to use the phase information to describe image texture. Firstly, a local variance operator is implemented on the amplitude and phase respectively. Secondly, the circular normal distribution and generalized Gaussian distribution are adopted to describe the phase and amplitude. Finally, the texture eigenvectors are composed of the parameters of statistic models. The results of texture classification experiments show that the proposed method can describe texture effectively, and its texture discriminability is better than the traditional methods based on DWT.Based on the spatial features of texture image, the statistical method of texture defect detection is studied using the one-class classifier. Support vector data description (SVDD) is a robust one-class classification method. However, its performance is strongly under the influence of kernel parameter selection. A novel parameter-optimizing method is proposed based on the volume duty ratio (VDR) of the one-class classifier. VDR can effectively estimate the degree for the SVDD classifier's boundary to be closes to the distribution of object sample. Several texture defect detection experiments on statistical texture surfaces, such as sandpaper, tile, casting, are implemented based on the one-class classifier. The results of experiments show that the suitable kernel parameter of a classifier can be selected by the proposed method, and the one-class classifier can detect the local defect in texture images only using the normal sample.The approach of texture defect detection is studied based on the frequency-domain characteristic of texture image. A novel method using DT-CWT reconstruction is presented for the inspection of local defects embedded in homogeneously textured surfaces. By properly selecting the smooth subimages and the detail subimages at different resolution levels for image reconstruction, the global repetitive texture pattern can be effectively removed and only local anomalies are preserved in the restored image. This converts the difficult defect detection in complicated textured images into a simple binary thresholding in non-textured images. Experimental results show that the presented method is suitable for the anisotropic (or regular) texture defect detection, and its performance excels DWT's.The vision inspection approach of aperture's radial crack on an aero-engine labyrinth disc is studied based on industrial videoscope. The circle detection and the radial crack recognition methods are two key components in the inspection algorithm. In order to improve the speed of circle detection and meet accuracy, a new fuzzy fast Hough transform (FFHT) algorithm is proposed. In the FFHT algorithm, the dimension of the parameter space is reduced using the local gradient information. In the parameter space, the coarse-to-fine search technique is used to reduce the computing and storage requirements of the hough transform. The experimental results show that FFHT can detect the circular contour quickly and accurately. Moreover, the fuzzy vote technique is adopted to lessen the uncertainty that arisen from edge pixel position error and gradient direction error. Moreover, two methods of aperture's radial crack detection are proposed based on chain code analysis and texture analysis respectively. The experiment of crack detection is carried out in aperture images, and the results indicate that the detecting accuracy using chain code analysis is 91.2%, and the accuracy with texture analysis is 95%. The texture analysis method has raised the accuracy of detection using edge density and direction in the local region synthetically.For inspection the defects on the surface defect of the rivet of aircraft frame and the control wire rope, which can not close easily and have complicated surface pattern, several texture defect detecting algorithms are designed and implemented correspondingly with the detecting mode that is "object detection→region of interest→defect identification". To detect the surface defect of the rivet and the control wire rope respectively. The LWS and Rodan transform are used to describe the texture feature of rivet regions, and the k-NN one-class classifier is adopted to identify the abnormality of rivet. In the inspection of the wire rope, wire rope images are segmented by texture feature firstly, then, the method based uniformity of subregion and the method using DT-CWT reconstruction are adopted to detect the surface defects of wire rope. Moreover, a nondestructive inspection system based on machine vision is developed. The results of detection experiments shows that the vision inspection methods based on texture analysis are effective to detect the surface defects of the rivet and the wire rope. These research results above show that the machine vision technology could be used widely in the nondestructive inspection field. The shortcomings of visual inspection with human eyes could be overcome by the machine vision detection technology. Simultaneously, the effective coverage of detection could be extended, and the accuracy and automation level could be improved.
Keywords/Search Tags:machine vision inspection, texture feature description, texture defect detection, local Walsh spectrum, one-class classification, dual-tree complex wavelet transform
PDF Full Text Request
Related items