本文讨论了一种基于亮度对立体图对左右图象的视差进行估计的方法。

In this paper an intensity based meth od to estimate the parallax between the left and the right image of a stereo pair is studied. Since the simplest intensity based method, i.e. block matching, is sensitive to noise, when small block sizes are required (as e.g. for computer vision purposes), a new method, improved block matching, based on the non linear diffusion process, was recently developed. It was shown that this new method has better noise suppressing properties than traditional block matching. Here, it is investigated how the new method reacts to a difference in average luminance and contrast of the left and the right image, caused by a gain difference in the left and the right camera. First, it is shown that improved block matching is less sensitive to these differences than traditional block matching. Secondly, a correcting mechanism, called gain control, based on the statistics of the left and the right image of the stereo pair, is shown to improve the performance of the parallax estimating algorithm.