A proposed stereo matching algorithm for noisy sets of color images

Modeling the world in three dimensions has been attracting a growing interest both in applications and science. In many cases, such 3D models are achieved by triangulating corresponding features recorded by several images of the same field taken from different points of view. This, however, requires the ability to match corresponding image elements detected in different images. In this paper, an algorithm for stereo matching in noisy pairs of outdoor images is described. The proposed algorithm applies a standard window-based correlation, but uses a fuzzy logic-based similarity function that models the HSV color space. This fuzzy logic modeling allows a robust color comparison that can tolerate a certain degree of changes in the illumination conditions and can be used for finding corresponding pixels in noisy sets of images. While the proposed algorithm does not introduce an improvement over existing methods in ideal conditions, experiments suggest that it provides significantly better results when the images in the set are relatively different from each other, and can be effective for matching corresponding features in images taken in different weather conditions, different positions of the sun, different optics or other real-life situations in which pinhole conditions are not available.