A fuzzy approach to solve the stereo correspondence problem using phase correlation

This thesis presents a fuzzy logic approach using phase correlation to solve the correspondence problem in stereo vision. The stereo correspondence problem is one of the fundamental and classical problems in computer vision. It involves identifying equivalent pixels between images of a given stereo image pair. The displacement between these pixels results in a disparity map, which relates to depth information of the scene. Until recently, the performance of most solutions degrades significantly when there is a change in illumination between the two images. Moreover, a high computational cost is required for a better performance. The primary aim of the proposed work is to improve the performance between images with possible lighting variations, while maintaining a low computational cost, to solve the stereo correspondence problem. To compute the disparity at a given point, the phase correlation technique is applied to windows at different horizontal positions on the same vertical image line. Using these results, a fuzzy logic inference system selects the most similar window to compute the disparity at the given position. Eight stereo image pairs from a well-known database [6,28] are used to perform experiments (see appendix 1.2), as well as images taken with our set-up (see appendix 1.5). For the case of the database images, errors with respect to the ground truth disparity maps and computation time for different window sizes are shown. The images from our set-up are evaluated with respect to real object depths. Utilizing both type of images, database and ours, it is shown that the proposed solution performs well, while maintaining a low computational cost, in images with lighting variations