基于改进的Graph Cut算法的羊体图像分割

Sheep image segmentation based on proposed graph cut algorithm

针对羊体图像复杂背景、不均匀光照且含有大量噪声等特点,提出一种融合多尺度分水岭的改进Graph Cut分割模型.引入多尺度分水岭对图像进行预分割,将基于像素级的Graph Cut算法转化为基于区域的算法以提高分割的效率.通过标记前景和背景种子点,利用模糊C均值算法实现前景和背景区域聚类.将多尺度分水岭分割的区域作为图割的顶点,以Lazy Snapping为框架计算图的边界项和数据项,并构造能量函数,通过最大流/最小割算法求解能量函数的最小值,从而实现图像分割.通过使用不同的分割算法进行实验比较,结果表明改进的算法在准确性和高效性方面都具有很好的性能.

Sheep images possess the characteristics of complex background,uneven illumination and much noise.The improved graph cut segmentation model combined with multi-scale watershed was proposed. Multi-scale watershed was applied to the pre-segmented regions instead of image pixels,so as to improve the efficiency of segmentation.Via labeling the foreground seeds and background seeds,foreground regions and background regions were clustered using the fuzzy C-means.The regions segmented by multi-scale watershed were regarded as the vertexes of graph,and energy function was built through computing the boundary term and data term based on lazy snapping. Max-flow/min-cut was used to calculate the minimum of energy function. Through the comparison of different segmentation algorithm experiments,results demonstrate the superior performance of the proposed method in terms of segmentation accuracy and computation efficiency.