结合CS-LBP纹理特征的快速图割算法

Fast Image Segmentation Algorithm Combining CS-LBP Texture Features

图割算法是目前最有效的交互式图像分割方法之一。针对当前景和背景颜色相似时容易发生分割错误并产生shrinking bias现象,以及基于像素的计算导致交互效率不高的问题,提出一种结合纹理特征的改进算法。该算法首先利用Mean Shift算法对图像进行预分割,构建区域邻接图,然后用累计直方图、CS-LBP纹理描述子对每个区域进行颜色和纹理特征的提取,通过在能量函数中引入纹理约束项以及局部自适应的正则化参数,有效改善了分割效果和shrinking bias现象。实验结果表明,本算法交互效率得到了提高,分割结果更加精确。

Graph cuts algorithm is one of the most effective interactive image segmentation methods. But it is prone to produce segmentation errors and shrinking bias phenomena when the color of foreground and background is similar and its interaction efficiency is not high due to pixel-based calculation. To improve these problems, an algorithm combining CS-LBP texture features was proposed in this paper. First the mean shift algorithm is applied to pre-segment the origi- nal image into regions to construct region adjacency graph. Then cumulative histogram and CS-LBP texture descriptor are used to extract color and texture features form each region. A new term of texture constraint is added to the energy function and local adaptive regularization parameter is used. So the segmentation effect and shrinking bias phenomenon are improved efficiently. The experiments show that interactivity efficiency and segmentation accuracy are improved.