A framework for HDR stereo matching using multi-exposed images

Abstract

Real world scenes that contain high dynamic range illumination present a special challenge for stereo matching algorithms due to a lack of texture in over- or under-exposed image regions. In this paper, we discuss possibilities for combining state-of-the-art stereo matching algorithms with High Dynamic Range (HDR) imaging techniques, in order to exploit a set of multi-exposed input images of both the left and right stereo view for high-quality stereo reconstruction. We sketch the overall concept of our HDR stereo matching framework and demonstrate some first steps of its implementation, including the acquisition of HDR stereo test data, stereo matching experiments on tone-mapped images, and ideas for combining disparity maps derived from different exposures.