Enhance support relation extraction accuracy using improvement of segmentation in RGB-D images

Todays, increasing in machine vision fields and applications make it necessary to have accurate scene understanding and analyzing. Support relation extraction is one of the most important and critical problem in robotic and machine vision task. In this article, we enhance support relation extraction accuracy using improvement of segmentation. Having the depth, moreover the color, in RGB-D images enable us to obtain accurate and precise support relation. In this paper an approach is also presented to redress discontinuities in point cloud occurred while recording. Experimental result shows the accuracy of the extracted support relation will be significantly increase after segmentation improvement.