采用全局相似性测量的彩色图像分割

本文提出一种新的采用全局相似性测量的彩色图像分割算法。该算法将图像分割问题表述为求解能量泛函最小化的问题。利用Bhattacharyya距离测量前景和背景之间的概率分布函数的全局相似性，并将Bhattacharyya系数作为最终的能量泛函。由于高阶能量项的引入，该能量的最小化通常是一个NP难题，为了能够有效的优化该能量泛函，本文提出一个辅助上界函数并且利用graph cuts对该函数进行优化求解。该辅助函数在优化的过程中能够保证能量递减。该算法能够应用于多种分割问题，包括交互式分割、显著性分割等。实验结果表明，算法具有全局分割的特点，能够对彩色图像进行较准确的分割。 

Color Image Segmentation by Using Global Similarity Measure

A novel color image segmentation algorithm by using global similarity measure is proposed in this paper. The algorithm states the image segmentation problem as energy functional minimization. The Bhattacharyya distance is utilized to measure the global similarity of probability distribution functions between foreground and background. The Bhattacharyya coefficient is adopted as final energy functional. Since the high order energy term is introduced, minimization of such energy is generally NP hard. In order to optimize the energy functional efficiently, an auxiliary upper bound function is proposed and it is optimized by graph cuts. The auxiliary function can guarantee to decrease the energy during the optimization process. The algorithm can be used in various segmentation problems, including interactive segmentation and sailency segmentation. Experimental results show that the proposed algorithm has global segmentation feature, which can segment color images correctly.