Shape-Based Image Retrieval Based on Improved Genetic Programming

Two-stage genetic programming algorithm based on a novel coding strategy (NTGP) is proposed in this paper, in which the generation of individual tree is not random but according to a special rule. This rule assigns each function operator a weight and the assignments of these weights based on the frequencies of function operators in good individuals. The greater weight of a function is, the more possibly it will be selected. By using the new coding strategy, the image feature database can be rebuilt. For two-stage genetic programming algorithm, in the first stage, the feature weight vector is obtained, GP is used to construct new features for the next step. While in the second stage, GP is used to induce an image matching function based on the features provided by the first stage. Based on these models, one can retrieve target images from the image database with much better performance. Three benchmark problems are used to validate performance of the proposed algorithm. Experimental results demonstrate that the proposed algorithm can obtain better performance.