Genetic Algorithm for Content Based Image Retrieval

In this work for CBIR system, all the image feature descriptors including color descriptors, texture descriptors and shape descriptors are used to represent low-level image features. Implementation of one feature descriptor doesn´t give sufficient retrieval accuracy. For combining of different types of features, there is a need to train these features with different weights to achieve good results. A real coded chromosome genetic algorithm (GA) and anyone performance evaluation parameter of CBIR like precision or recall are used as fitness function to optimize feature weights. Meanwhile, a real coded chromosome corresponding to higher precision as fitness function is used to select optimum weights of features. The optimal weights of features computed by GA have improved significantly all the evaluation measures including average precision and average recall for the combined features method.