Stereo image rectification based on polar transformation

Abstract

We propose a method based on a polar transformation between the source image and the output image for stereo image rectification. The proposed method solves several problems that arise from existing line-based methods. Many existing line-based methods require a different flow of rectifying process according to the different location of each epipole. Moreover, they inevitably require a recording table due to the lack of a transforming relationship. We solve these problems in a simple way by using a linear transformation that avoids the necessity of having different flows according to the locations of epipoles during rectification. Since all the epipolar lines use the same mathematical basis, look-up tables are not needed for referencing purposes. The proposed method consists of four steps, namely pretransform, common region detection, polar transform-based image rectifying, and epipolar line matching. The method is used on four pairs of stereo images obtained from three stereo structures. The results show good accuracy as well as good efficiency compared to existing methods in the literature.