Texture Description Of Micro-cell Image

Texture is one of the important characteristics of image. As an important research component of image understanding, analysis and recognition, texture analysis is in wide-range applications. Based on cell image, some approaches on texture analysis are introduced and compared. According to the features of cell image, some improved methods and new methods are put forward. These methods can describe texture of different kinds of cells. The main content in this paper is presented as follows:1. Improved synthesized Gray level co-occurrence matrix. Based on low resolution micro-cell image, this paper points out some problems when traditional Gray level co-occurrence matrix is used in texture analysis. The synthesized Gray level co-occurrence matrix can be calculated through weighted coefficient of different directions, texture characteristic parameters of which will be used to describe texture characteristics of cells. On analyzing correlations between texture characteristic parameters, redundant ones are dropped and four of them are chosen finally. The influence of distance on texture characteristic parameters is discussed, and the proper distance is determined at last.2. Based on 20-resolution micro-cell image, topology analysis method is proposed. According to the topology structure of inner cell, threshold segmentation method and operator segmentation method are used to get threshold image. Sobel operator, Prewitt operator and Laplacian operator are chosen in operator method. Conditioning region-growing is used to detect regions in threshold image. Topology characteristics are then calculated.3. Based on crystal in 20-resolution micro-cell image, moulding-board matching method is proposed. In view of outer shape and inner texture characteristics of crystal, the relevant mathematical model is established. The rotating angle is calculated based on two values, one is extremums of cell in the direction of x and y axis, the other is pels number of cell. With the angle, matching moulding board and matching rate can be got.4. The texture analysis methods that have just been mentioned are programmed with VC++. The performance of these methods are analyzed and the rationality of them are validated.