CVPRW '04 Proceedings of the 2004 Conference on Computer Vision and Pattern Recognition Workshop (CVPRW'04) Volume 12 - Volume 12
| Hi-index | 0.00 |
A Genetic Algorithm for graphical pattern matching based on angle matching had been proposed. It has proven quite effective in matching simple patterns. However, the algorithm needs some modifications to enhance its accuracy on pattern matching when there are some differences between two patterns in terms of numbers of nodes, shapes and rotations. This paper presents the modifications, such as the introduction of node exemption, inexact matching between straight lines and curves in the patterns, and consideration of rotational degrees of the patterns. Each angle is also given with a weight to indicate the significant degree of the angle. A multi-objective function is used to reflect the similarity between two patterns. The experiments designed to evaluate the algorithm have shown very promising results. It is highly accurate on patterns matching with dissimilarities in shapes, numbers of nodes and rotational degrees.