Study On Color Image Segmentation Algorithm With Depth Information

Image segmentation refers to all the pixels of an image based on an attribute, according to certain criteria, to be divided to form a plurality of non-overlapping and cover the entire field area. Traditional image segmentation techniques divide the image into a number of areas with the appearance of image feature.But this algorithms can’t distinguish the foreground and background which with the similar appearance of image feature, and that will load to the over segmentation and under segmentation. And the robustness of the segmentation will be poor cause of light, shadow and other effects.Fusion depth information for image segmentation can well overcome the above shortcomings. The value of depth image pixels shows the distance of camera to object surface, and it’s independence to color. The depth image can be not affected by the impact of factors such as illumination and shadow, at the same time depth image is very sensitive to object distance and can distinguish the foreground and background objects easily.This paper studies and achieves Kinect for color image segmentation algorithm based on the depth of integration of information. This algorithm divide images with the color superpixel segmentation of color images and the information of Kinect depth images. Aiming at the problem that the depth image of Kinect is not accurate, noise and with holes, we give a precise calibration to Kinect, and get the relationship between the Kinect color camera and the depth camera. We align the color image and depth image to fill the holes with the consistency of color feature. Then we proposed a modified bilateral filtering to get a fine and smooth depth image.In this paper, we study the SLIC superpixel segmentation method based on color information, and analyse the results of SLIC segmentation. For the phenomenon of error segmentation and over-segmentation, the following two algorithms are adopted. One is to use the depth information to identify the SLIC error segmentation and make the OTSU algorithm to fix the error segmentation; Second, in view of the SLIC over-segmentation phenomenon, this paper puts forward that a super pixels merging algorithm based on color and depth information, make the color and the depth information as superpixel features, and then to ultra pixels redistributing labels, distributing the features similar to pixel block the same label, implements the effective merger pixel block.