基于多种边缘暗示和尺度修正的RGB-D图像层次分割

Hierarchical segmentation for RGB-D images based on contour cues and scale alignment

针对现有RGB-D图像分割算法中存在的过分割和欠分割问题,提出一种基于多边缘暗示和尺度修正的层次分割算法。首先,利用引导滤波对RGB图像去噪;然后,结合RGB-D图像中的单目边缘暗示和深度图像中的几何边缘暗示,进行边缘检测,并将边缘梯度作为初始边缘强度,利用有向分水岭变换,得到超度量轮廓映射,通过控制阈值,获取层次分割;最后,对不同层次的区域分割进行尺度修正。实验结果表明:通过结合多种边缘暗示后的边缘检测算法能够有效提高区域轮廓的平均准确率,而经过尺度修正后的层次分割算法能够有效提高区域的最佳覆盖率。

A novel method is proposed by utilizing multiple contour cues and scale alignment,to help better the balance between over segmentation and under segmentaton for hierarchical RGB-D image segmentation. Firstly,the guided filter is used for dealing with the noise of RGB image. Then,monocular cues are combined with geometric cues of RGB-D image for contour detection and the gained gradients of the boundary are used as the initial strength of the boundary for oriented watershed transform to construct the ultrametric contour map,thus building a hierarchical segmentation by controlling thresholds. Finally,the scales of the regions at different levels of the hierarchy are modified. Experimental results show that the method can improve the average precision of the contour and best covering of the regions after taking advantage of mixed contour cues and scale alignment.