T exture Segmentation Model Driven by the Simplified Cauchy-Schwarz Divergence

This paper addresses a new segmentation model for texture image based on active contour and the model is driven by the simplified Cauchy-Schwarz divergence. First, we define a new texture descriptor with the Gauss curvature and the mean curvature of the textural region to represent the textural image space. Then, a novel active contour model has been developed to distinguish between object of interest and background by computing the simplified Cauchy-Schwarz divergence of the probability density functions. Finally, an alternative minimization procedure and the Chambolle’s fast duality projection algorithm are adopted in order to efficiently solve this model. Experimental results show that our model has better segmentation performance than the traditional models driven by the Kullback-Leiber divergence and the Bhattacharyya distance for both synthetic and natural texture image.