DSLIC: A Superpixel Based Segmentation Algorithm for Depth Image

Limited illumination outdoor and indoor environment leads to the lack of object's color information. Faced with this situation, it is not always possible to generate superpixel by using RGB or LaB features. To tackle this scenario, we propose a superpixel generation algorithm solely on depth image. We aim the semantically-incoherent superpixel problem on depth image, caused by identical depth value in the vicinity of the border. Our algorithm is an adaptation of Simple Linear Iterative Clustering (SLIC) with a novel utilization of depth and gradient direction as an alternate of LaB color space features. Our novel approach is demonstrated perform favorably to over-segment large planar area in an unlit environment.