Dynamic programming-based dense stereo matching improvement using an efficient search space reduction technique

Two most important evaluation parameters of any stereo matching algorithm are the execution speed and the accuracy of generated results. Commonly, reported real-time works provide fast depth calculation based on highly specialized hardware. Thus, algorithmic techniques for increasing the speed which can be employed on every general hardware, are more valuable. In this paper an efficient technique has been proposed in order to improve dynamic programming (dp) based dense stereo matching. The main idea is using a fast and efficient search space reduction algorithm and successively forcing dp based matching process to operate on this reduced search space (RSS). Effective ideas have been proposed to correct RSS and emphasize on more qualified candidates in RSS to facilitate and improve matching process in this space. Experimental results on standard stereo images show significant increasing of execution speed as well as error reduction for two traditional and basic versions of dp based matching algorithms. These simultaneous improvements in both speed and accuracy, considering their usual inverse relationship in other algorithms, seems to be very valuable.