Study On Feature Extraction And Classification For Color Texture Image

Image classification holds a central position in the field of computer vision and pattern recognition. As the foundation and precondition of classification, feature extraction directly influences the performance of the classifier. Texture characteristic is basic feature of image which is playing an important role in image analysis. In addition to the texture information, color characteristic is also essential for color texture image, how to combine the two information effectively is worth thinking about.Gray level co-occurrence matrix(GLCM) and local binary pattern(LBP) are two of the common texture feature extraction methods based on statistics, especially the local binary pattern method, with its simple theory, high computational efficiency, etc, has been widely used in texture analysis. But LBP is sensitive to light and noise, and in the real world, the change of illumination and noise is very common in color images.For the problem of illumination change, based on opponent color theory, this paper proposes a new method of extracting texture feature, called Local Oppugnant Color Extrema Patterns In Single Color Space(SCS_LOCEP). This method strives to use the harmonized link between texture and color, and extracts the directional texture information based on local extrema in 0,45,90? ? ?and 135?directions from every oppugnant channel in RGB space.Finally, It's combined with magnitude patterns of every oppugnant channel as feature of color texture image.Because pixels of some sub-block images are larger difference in every channel of RGB space, it comes the problem of pseudo noise pixels in some sub-block images of oppugnant channel. According to this problem, this paper also puts forward using adaptive window to choose appropriate threshold to improve the SCS_LOCEP. Series of the experiment results in SVM classier show that improved method has a better classification accuracy, and have good robustness for light and noise.