Efficient image segmentation of RGB-D images

Abstract

Image segmentation is a fundamental problem in computer vision. With the current advent of depth sensors, it is gradually becoming a research focus on how to utilize the depth information to improve image segmentation. This paper proposes an automatic RGB-D image segmentation method in which the depth and RGB images are separately segmented and the result is combined, hence obtaining better segmentation results. The proposed segmentation is applied in five phases: 1) Edge detection, 2) Morphological operations employed for enhancing the edge detection result. 3) Connected components' processing applied for labeling each region in the image, 4) Extraction for the missing components and merging with result in step 3. (The previous four steps are applied on the RGB image). 5) The result of depth and RGB segmentation are finally combined. Experiments carried on 'NYU Depth Dataset V2' which contains RGB and depth images, have proven the efficiency of the proposed segmentation method.