The USC Andrew and Erna Viterbi School of Engineering USC Signal and Image Processing Institute USC Ming Hsieh Department of Electrical and Computer Engineering University of Southern California

Technical Report USC-SIPI-124

“Stereo Matching Using a Neural Network”

by Yi-Tong Zhou and Rama Chellappa

March 1988

A method for matching stereo images using a neural network is presented. Usually, the measurement primitives used for stereo matching are the intensity values, edges and linear features. Conventional methods based on such primitives suffer from amplitude bias, edge sparsity and noise distortion. We first fit a polynomial to find a smooth continuous intensity function in a window and estimate the first order intensity derivatives. Combination of smoothing and differentiation results in a window operator which functions very similar to the human eye in detecting the intensity changes. To give some insights into the resulting window operator, a theoretical analysis of the variances of the estimated derivatives is given. Since natural stereo images are usually digitized for the implementation on a digital computer, we consider the effect of spatial quantization on the estimation of the derivatives from natural images. A neural network is then employed to implement the matching procedure under the epipolar, photometric and smoothness constraints based on the estimated first order derivatives. Owing to the dense intensity derivatives a dense array of disparities is generated with only a few iterations. This method does not require surface interpolation. Experimental results using random dot stereograms and natural images pairs are presented to demonstrate the efficacy of our method.

To download the report in PDF format click here: USC-SIPI-124.pdf (2.0Mb)