基于SLIC的GrabCut减小姿态搜索空间算法

GrabCut Algorithm Based on SLIC for Pose Search Space Reduction

针对人体的高自由度导致姿态估计过程中搜索空间过大的问题,提出一种基于简单线性迭代聚类(SLIC)超像素算法的Grab Cut减小姿态空间算法。运用SLIC算法对图像进行超像素分割,以超像素作为s-t图中的节点构建图模型,利用超像素区域的颜色特征平均值作为该区域内每个像素的特征值,分别为前景和背景超像素建立混合高斯模型,迭代更新高斯参数,运用最小割算法完成前景提取,并在得到的前景区域中进行后续的姿态估计。实验结果表明,基于SLIC的Grab Cut与基于Grab Cut的减小搜索空间算法在运行时间和姿态估计准确度上均有较大程度提升。

In order to solve the problem of extremely large size of pose search space due to body parts＇ high degree of freedom during pose estimation, a pose search space reducing algorithm of GrabCut based on Simple Linear Iterative Clustering （SLIC） superpixel approach is proposed. SLIC algorithm is used to segment images into superpixels which are applied as nodes to build a s-t graph. The mean value of color feature in the area of a superpixel is used as the feature value of each pixel in that area. Foreground and background Gaussian Mixture Models （GMM） are respectively built and Gaussian parameters are updated using iterative processing. Image foreground extraction is achieved using Min Cut. Pose estimation is performed only in the foreground area obtained by foreground extraction. Experimental results show that comparing to pose search space reduction method based only on GrabCut, the algorithm of GrabCut using SLIC has much better performance on both running time and pose estimation accuracy.