基于图模型及骨架信息的人体分割算法
Body Segmentation Algorithm Based on Graph Model and Skeleton Information

针对复杂场景中分割人体不准确的问题, 提出了一种在图论优化框架中联合RGB-D信息和骨架信息的人体分割算法．首先, 采用边缘引导的滤波算法修复低质量的深度图, 得到高质量的深度图; 然后通过一种聚类算法对RGB-D数据进行聚类得到超像素; 最后在图模型中将超像素看作节点, 并结合相应的人体骨架来提高区分人体和背景相似颜色区域的能力, 设计能量函数各组成项, 最小化能量函数得到全局最佳的融合结果．为验证算法的有效性, 在实际场景数据集上与多种算法进行比较．实验结果表明, 在主观视觉和客观指标上, 本文提出的算法均得到了更为准确的人体分割结果．
To resolve efficient human body segmentation from complicated background，a body segmentation algorithm based on the graph-based optimization framework with the combination of RGB-D and skeleton information was proposed in this paper. Firstly，an edge-guided filter algorithm was adopted to recover low quality depth map and obtain high quality depth map. Then the RGB-D data was clustered into superpixels via a clustering algorithm. Finally，a graph model was proposed，in which the superpixels were considered nodes and the associated skeleton was incorporated to enhance the capability of the graph in distinguishing body regions with similar color to the background. Each component of energy function was designed and optimal global merging result was obtained by minimizing the energy function. To evaluate the effectiveness of the proposed algorithm，several experiments on the real scenarios were compared in this paper. Experimental results show that the proposed method achieves more accurate body segmentation performance in both subjective visual and objective index comparisons