The Ceramic Tile Image Classification Based On Texture Feature

With the extensive application of tile and increasingly rich tile types, tile sorting seems in low efficiency when it comes to large quantities and multi-species production. Therefore, using machine vision technology is significant for intelligent tile classification. Existing ceramic tile classification system has been mostly only using the color feature of ceramic tile or simple texture edge information, ceramic tile classification these feature information is only applicable to monochromatic tiles or simple patterns, not suitable for ceramic tile texture classification. In response to this situation, we combine the gray level co-occurrence matrix and statistical geometrical method to do the research on hierarchical classification of texture analysis and classification algorithm on tiles with monochrome and no texture, simple regular pattern and complex irregular patterns, etc.Due to the light conditions and other factors in the work site, the captured tile images have the problem of uneven illumination and unobvious texture. To deal with the problem, differential high-pass filter and histogram equalization are applied to enhance the texture of the tile image. To solve the problem of changes of location and direction on the practical conveyor belt, GLCM with shift invariance and statistical geometric features methods with rotation invariance are used to extract the texture characteristics of tiles. Moreover, the Support Vector Machine (SVM) is adopted to build the classifier because of the limited training samples of tile images and feature space with nonlinear characteristics. This paper analyzes and compares different kernel functions on the accuracy of Support Vector Machine (SVM) classification, and choose K cross-validation method for parameter optimization of SUV parameters. The separate use of a texture analysis feature extraction method description ability is limited, the texture features and simple combining different methods get there may be caused by reduced feature redundancy problem classification recognition rate, this paper combined with different texture analysis method of layered classification.As the experiment shows, the methods of texture enhancement, feature extraction and classification in this paper can effectively realize the automatic classification of tile image.