结合彩色和空间信息的RGBD图像超像素分割算法

Superpixel Segmentation of RGBD Image Joining Color and Space Information

针对室内场景的特性,提出一种结合彩色和空间信息的简单线性迭代聚类(SLIC)算法来对室内场景RGBD图像进行超像素分割。该算法提出一种将彩色和空间信息相结合的方法来计算不同像素点之间的距离,具体地说,该算法利用两个像素点的法向量之间的余弦值对它们的彩色差异值和空间位置差异值进行自适应加权。该计算方法能有效区分室内场景中两个颜色相近的方向不同的相邻平面区域。在聚类过程中,聚类中心的搜索范围是固定不变的,整个聚类过程在线性时间内完成。在NYUv2图像数据集上的实验结果表明,对比只使用彩色图像的SLIC算法,本研究提出方法的分割性能更好,边缘定位更准确。

For indoor scenes,this paper proposes a superpixel segmentation method based on the simple iterative clustering(SLIC)algorithm with joining color and space information.Specifically,a method joining color and depth information is proposed to compute distance between two pixels,where consine value between two normal vectors of the two pixels is computed to weight differences of the color and the space position adaptively.The proposed method can successfully distinguish two adjacent planes of indoor scene with similar color but different normal direction.The whole clustering process can be finished in linear time since the searching range of each cluster is fixed.The segmentation results on NYUv2 dataset show that,comparing with the original SLIC algorithm with only color information,the proposed method can obtain better results with more accurate boundary localization.