Far East Journal of Experimental and Theoretical Artificial Intelligence
Volume 6, Issue 1-2, Pages 1 - 23
(November 2010)
|
|
INTELLIGENT STEREO MATCHING ALGORITHM BASED ON HOPFIELD NEURAL NETWORK AND GENETIC ALGORITHM
Shih-Hung Yang, Cheng-Yu Ho and Yon-Ping Chen
|
Abstract:
This paper presents intelligent stereo matching algorithm (ISMA) to solve problems associated with matching stereo images. This algorithm adopts a two-dimensional Hopfield neural network (HNN) to match stereo pairs based on an energy function including three constraints referred to as uniqueness, similarity and compatibility. The similarity of a matched pair is obtained by identifying differences in the neighborhood around the matched feature points. The compatibility of a matched pair to neighboring matched pairs is calculated via not only smoothness and geometric comparisons, but also vertical disparity comparison to enhance the correct matching percentage. Furthermore, ISMA employs a genetic algorithm which systematically determines the weights and parameters of the energy function. In addition, ISMA applies an updating rule enabling HNN to efficiently identify correct matched pairs. Finally, the experiment results obtained from a binocular robot demonstrate the applicability of ISMA. |
| Keywords and phrases: stereo matching, Hopfield neural network, genetic algorithm, vertical disparity, similarity. |
| Communicated by Shun-Feng Su |
| Number of Downloads: 174 | Number of Views: 394 |
|
