Color Texture Image Segmentation Using Histogram-Based CV Model Driven by Local Contrast Pattern.

This paper proposes a novel texture descriptor called local contrast pattern (LCP) to drive a histogram-based Chan-Vese(CV) model for color texture image segmentation. The LCP has two features: differential contrast and orientation. The former measures the variation of local intensity and the latter extracts the texture orientation information. In order to enhance the localization, a truncated Gaussian kernel function is also incorporated that determines a positive correlation between weight and distance. Then, a novel histogram-based CV model is established which is guided by a combination of the LCP feature maps and the kernel to obtain color texture segmentation. The effectiveness of the LCP descriptor in color texture segmentation was prove to be true by many single variable validation experiments. Comparisons with many color-texture image segmentation models demonstrate that our proposed model can not only successfully partition all types of images but also bear strong robustness for illumination, noise and curve initialization.