Multi-scale stereo matching using edge dynamic programming
Jun Xiao · Cehui kexue · 2012
For the weakness of the pixel-based stereo matching,setting segment as a processing element is more practical.This paper presented a stereo matching method using an object-oriented multi-scale method to achieve from coarse to fine stereo matching.Firstly,from the perspective of human stereo vision,the reference image was segmented by an object-oriented segmentation method with large scale,which simulates human perceptual process and fully takes shape,spectral and spatial context into account.Secondly,by employing a refined D Palgorithm which is based on the inner edgeline of the segment,the segment's disparity was obtained.Thirdly,according to the accuracy requirements,the segment was determined whether need to be further subdivided to obtain more precise disparity map.Finally,the final disparity image was obtained.The experimental results indicated that consistent with human being's vision,this algorithm combining segmentation with refined D Palgorithm sped up stereo matching process while keeping matching results accurate.