中文摘要：

针对传统Grab Cut在GMM迭代估计阶段仅单纯地考虑像素点的RGB彩色信息,当前景细节区域与它的周围区域颜色差异较大时容易发生分割错误,以及基于像素的运算导致分割效率不高的问题,提出一种结合权值优化与CS-LBP纹理特征的改进算法。该算法利用多尺度分水岭对图像进行预分割,构建区域邻接图;然后对每个区域进行颜色和纹理特征的提取,通过权值迭代优化算法使区域的数据项权值与周边分块区域的权值相关联,采用自适应参数将纹理约束项引入能量函数,并将改进算法应用于人脸图像分割,有效改善了分割效果。实验结果表明,该算法分割结果更加准确,效率更高。

英文摘要：

The stage of estimating the GMM iteratively only considers the pixels＇ RGB color and the calculation is based on pixels in traditional GrabCut, so it is prone to produce segmentation errors when the details of foreground and its surroundings are different and its efficiency is not high. To improve these problems, this paper proposed an algorithm combining the optimization of weights and CS-LBP texture feature. Firstly, this algorithm applied the multi-scale watershed to pre-segment the original image into regions to construct region adjacency graph. Then it extracted color and texture feature from each region and used the optimized algorithm iteratively to correlate the weight of data item of a region with its surroundings. The energy function added the texture constraint with adaptive parameter and applied the improved algorithm to facial image segmentation. The algorithm improved the segmentation effect efficiently. The experiments show that the proposed method can improve segmenta- tion accuracy and efficiency.