Image Segmentation by Fusing Color and Depth Information for Region Merging

When there are occlusions and shadows in the image, it is difficult for the traditional algorithm to extract the object contour accurately by using only the appearance information. In this paper, an improved Mean-shift algorithm is used to obtain the initial set of oversegmentation subregions. Then the color contrast of the subregions is calculated, the depth contrast is calculated in the depth image, and the fusion contrast is obtained by a dynamic weighted curve fitting method. Finally, the seed regions of the target and background are automatically selected according to the depth information, and MSRM algorithm is used to complete the image segmentation based on the principle of maximum similarity. Experiments on NYU-V2 and Middlebury database show that compared with the current general algorithm, the proposed algorithm can improve the accuracy of segmentation and improve the visual effect of segmentation images.

I. Introduction

Image segmentation is a widely studied subject in the field of computer vision, which has been widely used in object recognition, scene understanding, robot navigation and other fields^[1]-[2]. Objects are defined by lower-level features such as color and texture, as well as physical connectivity, which is represented by the continuity of scene depth in three-dimensional space^[3]. As an additional feature, depth information can improve the accuracy of image segmentation. For example, when the color of the target and the background are similar, or there are low-contrast edges in the image, the depth information can be used to correctly distinguish each target object and the background.