Segmentation of color textile images based on a multiscale context model

In this paper, a multiscale color image segmentation algorithm based on a contextual model for textile images corrupted by textile texture noise is proposed. The context model not only captures the statistical dependency between adjacent scales, but also the statistical dependencies among the neighboring blocks. The multiscale approach is used to solve the conflict between boundary localization and high resolution segmentation, and the segmentation result is recursively refined at each scale based on the contextual model. Experimental results show that our algorithm can achieve better segmentation results when tested on color textile images, and also produce high quality edge images when compared with the mean-shift algorithm and the multiscale block segmentation approach.