Superpixel segmentation is to gather adjacent pixels into sub-regions with certain semantics, which can greatly reduce the computational complexity of subsequent image processing. The existing SLIC-D algorithm of RGBD image is a superpixel segmentation method that integrates depth information based on SLIC algorithm. However, for some images with complex background, it cannot stick to the weak edge of the image, and even has the defect of wrong segmentation. Therefore, this paper proposes a novel RGBD image superpixel segmentation intergrated depth map quality. Firstly, RGB map is enhanced by bilateral filtering to suppress and smooth noise, joint bilateral filtering is used to fill the depth map hole. Secondly, Holistically-Nested Edge Detection method based on convolutional neural network is used to obtain the RGB contour map, the edge consistency map is obtained by Hadamard product operation with depth gradient map. Finally, superpixel segmentation is performed by clustering based on LAB color space, spatial location and high-quality depth information similarity. Through the verification on NYU dataset, the advantages of this algorithm in edge fit degree are fully demonstrated.