Combining Multiple Images to Improve PatchMatch Stereo

Abstract

Patchmatch based stereo is an efficient method for stereo computation, and also has been extent to multiple-view stereo recently. In this paper we extend the two-view patchmatch stereo to multiple-views in the multiple-view stereo pipeline. The key of the proposed method is to first select multiple suitable neighboring images for a reference image and then combine them under the patchmatch stereo framework. Experimental results on benchmark data sets demonstrate the accuracy and efficiency of the proposed method.