An Accurate Stereo Matching Method with Chaos Genetic Algorithm Using Rank Transform and Morphology

Stereo matching plays an important role in the machine vision. One of the most important steps in the stereo vision process is to finding corresponding image points from two view of scene. This step called the stereo matching. In this paper, we propose a new method for generating accurate disparity map based on genetic algorithm. In our approach, the stereo matching problem is considered as an optimization problem. Genetic algorithms are efficient search methods based on principle of population genetics, i.e. mating, chromosome crossover, gene mutation, and natural selection. GA with excellent capabilities solves difficult nonlinear optimization problems. According to the phenomenon of chaos, the chaos operation will be added the searching process for the genetic algorithm in this paper, so can improve the performance of optimization procedure. In chaos, a small difference in the initial conditions may produce an enormous error in the final phenomena. The rank transform is applied on left and right images firstly. Rank transform is a form of non-parametric local transform which has been used for stereo matching problem. This technique is used to rank the intensity value replaced by its rank amongst the neighboring pixels. Then we used the morphology to achieve the disparity map with less discontinuity. Mathematical morphology is a set of algebra used to process and analyze data based on geometric shapes. In this article, we used the morphological technique for post processing. We use RMSE to compare our results with other approaches. The Root Mean Square Error (RMSE) is a frequently used measure of differences between values predicted by a model or an estimator and the values actually observed. RMSE is a good measure of accuracy. In our experiments, we used image pairs Tsukuba to test the proposed algorithm. In order to testify the performance of our algorithm, the simulation result was compared with those got by a fixed-size window. The experimental results show that our approach can generate more accurate disparity maps than two existing approaches.