Texture Restricting Flexible Band Active Contours for Segmenting Desired Object in Color Images
Tengyue Zou, Xiaoqi Tang, Bao Yun Song, Jin Wang · 2012
In this paper, we propose a flexible band variable Chan-Vese model to segment the desired object in color images. Classical Chan-Vese model based on level set implementation can handle topological changes and detect the interior and exterior boundaries of all objects automatically, but cannot segment the desired object alone. We build the texture restricting probability distribution from color and texture features to recognise the desired object. The color feature is extracted by running the back projection operation on the Hue Saturation Value color system. The nonlinear structure tensor is applied to get the texture features. We further introduce flexible band region energy to reduce computational consumption and improve segmentation accuracy. Evolution of deformable curve is performed by energy minimization on the level set framework. Experimental results indicate that our approach obtains more accuracy with less time cost in segmentation.