结合显著性检测和图割的RGBD图像共分割算法

RGBD Image Co-segmentation via Saliency Detection and Graph Cut

针对前背景颜色相近的图像难以准确分割的问题,提出了一种结合显著性检测和图割的RGBD图像共分割算法。不仅实现了对多幅图像的共分割,还借助深度数据来解决前背景混淆的问题。将深度引入超像素分割算法,每幅RGBD图像变成超像素块的集合;构建超像素块的图模型,结合显著性检测来扩充种子点区域,基于Biased Normalized Cuts来实现共分割;借助深度信息来优化分割结果。实验表明:对于前背景颜色相近的RGBD场景,能显著提高分割结果的准确度。

In order to solve the problem that it is difficult to accurately segment images with similar colors in the foreground, we propose a RGBD image co-segmentation algorithm that utilizes saliency detection and graph cut. Our algorithm not only achieves the co-segmentation of multiple images, but also uses depth data to solve the foreground and background confusion problem caused by color similarity. Depth is incorporated into a superpixel segmentation algorithm to change each RGBD image into a set of superpixel blocks. A graph model of superpixels is constructed and saliency detection is used to extend the seed nodes area. The co-segmentation is achieved based on the Biased Normalized Cuts. Depth information is used to further optimize the segmentation results. Extensive experiments show that our method can significantly improve the accuracy of segmentation for those scenes with similar foreground and background colors.