Multi-channel with RBF neural network aggregation based on disparity space for color image stereo matching

Abstract

Stereo matching is widely using for 3D reconstruction, which aims to obtain corresponding locations between pairs of stereo images. In this paper we present a robust neural aggregation method for matching correspondences in stereoscopic color image. A data structure disparity space image (DSI) was firstly introduced for development of a local-based matching algorithm. To make good use of color information, stereo images were represented by RGB model, and the initial disparity dense map of correspond RGB channels were computed using NCC (normalized cross-correlation) based on DSI matching algorithm. The neural network performed the similarity aggregation of RGB channels, and the aggregated method shown not only a better overall behavior, but also the neural will improve the robustness of area-based matching methods which depend on the proper selection of window shape and size. The experimental analysis makes a comparison with other methods that show neural aggregation with more matching accuracy.