Neural disparity map estimation from stereo image

Abstract EN

In this paper, we propose a new approach of dense disparity map computing based on the neural network from pair of stereo images.

Our approach divides the disparity map computing into two main steps.

The first one deals with computing the initial disparity map using a neuronal method Back-Propagation (BP).

The BP network, using differential features as input training data can learn the functional relationship between differential features and the matching degree.

Whereas, the second one presents a very simple and fast method to refine the initial disparity map by using image segmentation so an accurate result can be acquired.

Experimental results on real data sets were conducted for evaluating the neural model proposed.