Improvement of Target Extraction and Dense Matching

Abstract

In order to obtain dense 3D point clouds of the target based on image sequence, it is necessary to ensure the accuracy of the target extraction and the dense of the disparity which based on the dense matching of image sequence target. A new method that was used to improve the target extraction and dense matching was proposed in this paper. In the image target extraction, many interactive iteration segmentations method was adopted to amend the inaccurate region that was segmented by traditional GrabCut; and in the dense matching, the image texture features information was introduced to divide the image into the texture-rich region and sparse texture region, and different dense matching methods were applied to the different regions. This method solved the problem that image target can not be accurately extracted from a complex scene; On the other hand, the localized reconstruction due to the fact that image texture sparse region can not be achieved quasi-dense matching was also solved. The results displayed that, after the using of this method, good results have been showed, and the accuracy of three-dimensional reconstruction was improved.