An Improved Fusion Algorithm of the Weighted Features and its Application in Image Retrieval

Abstract

In order to improve the average recall rate and the average precision rate of image retrieval, an improved fusion algorithm of the weighted features is presented. Firstly,the shape features of images are extracted by using the moment invariant method based on 7 central moments. Meanwhile,the texture features of images are calculated by using the Gray-level Co-occurrence matrix. Then the elements of the vectors are normalized respectively. In the next step, the Euclidian distance,the squared Euclidian distance and the City-Block distance are calculated. The Mean values of the 3 kinds of distances are obtained and used as the shape distance and the texture distance. Finally,the weighted feature vectors are fused and the similarities between images are obtained and used as the measure bases to implement the image retrieval. The experiments show that the tangible results of image retrieval are realized and the average recall rate and the average precision rate are improved.