Texture Analysis Based Object Recognition Technology Research And Development

Visual activity is the main way for people to understand of the outside world. The main visual perception come from the object's shape, color and texture. Texture are widely exists in nature as a fundamental attribute, it is a very important characteristic of describing and recognizing the objects. Texture analysis is one the most important function of perception for human. In machine vision, texture research's main purpose is to understand, modeling and processing of image texture pattern, with the ultimate goal of intelligent recognition. Texture analysis based target recognition has been widely used in industrial products, food safety detection management, document classification, financial forecasting, multimedia database retrieval, biomedical, and national defense security. The focus of this study for texture analysis technique in textile defect target recognition application, mainly includes the following aspects:(1) Summarize the commonly used image texture description methods, which can be divided into two categories, the first category is based on the time-frequency decomposition of image texture description method. The second category is the texture image is divided into portions of the structure and texture of some image processing operations.(2) Studied the function of wavelet transform used for the singular signal detection on the periodic texture image, and proved by experiments that the wavelet transform is the ideal tool of identifying fabric defects.(3) Studied the wavelet parameter expressions, get wavelet parameter equation which including different parameters, proposed search parameters of the parameter equation to realize the adaptive signal wavelet decomposition. According to the need to define the fitness function, use improved genetic algorithm to search the best wavelet. Using this algorithm to enhance the defect signal of fabric, and compared with common stationary wavelets decomposition results, the results show that the adaptive wavelet can more fully isolated plain cloth flaws in the information.(4) Presents a method based on texture feature to defect monochromatic plain weave fabric, and used for the on-line detection of warp knitting machine. In order to meet the needs of online detection speed requirements, design off-line learning, on-line detection scheme. In offline stage by learning the normal cloth for the adaptive wavelet, texture cycle as well as the segmentation threshold parameter; on-line detection phase, firstly monolayer decompose fabric image by adaptive wavelet, and then extract the image texture features, finally through the adaptive double threshold segmentation to recognize flaw signal. At the same time, the detection system hardware equipment are described in detail.(5) Study the complex patterned texture description method, proposed a watershed segmentation based pattern texture fabric defect detection method. The method uses regular band as pattern texture features, and then using the watershed algorithm to defect segmentation. And compared with the proposed method, this method is more simple, more robust.