Incorporating Angular Ratio Images into Two-Frame Stereo Algorithms

Abstract

Abstract. Light Transport Constancy (LTC) asserts that the reflectance ratio obtained from two different illumination variations remains constant for any given view of the observed scene. LTC was proposed in [21] as a rank constraint for solving the correspondence problem in multiple view stereo. In two-frame stereo, the simplest setting for LTC requires only two illumination variations and a single light source. Under this scenario, the rank constraint can be formulated through ratio images, and standard stereo algorithms can be applied in order to obtain a disparity map. Unfortunately, a ratio image may be subject to saturated pixel values, and this may diminish the quality of disparity maps. To solve this problem, as a first contribution in this work, we propose a post-processing operation based on slope angles related to the ratio values. Experiments show that new angular ratio images are more robust and deliver improved disparity maps. A second contribution of this paper consists in performing an experimental evaluation of angular ratio images under the standard test bed for two-view stereo algorithms, i.e., using different aggregation and optimization approaches. The results of our research are consistent with previously reported conclusions for two-view stereo surveys. It means that LTC may benefit from a vast variety of existent methods to solve the twoview stereo problem.