Design of a Fuzzy Distance for a CBIR System
Oscar D. Robles-Sánchez, Enrique Frı́as-Martı́nez · 2002
This paper describes the design of a fuzzy distance for content-based image retrieval applied to 2D color images within an image database. The technique herein presented computes first a multiresolution representation using the Haar transform. Then, a color based primitive is obtained from the color level energy distribution at different resolution levels of the wavelet transform. Fuzzy membership functions are constructed from the values of the color based primitive of the test vector. The values of the elements of the database images are used to compute a fuzzy distance between the test and the database images. The experimental results given in this paper include the success rate achieved with the proposed distance using different membership functions and aggregation operators over a database composed of 298 color images and 68 tomographic volumes with test sets formed respectively by 1655 images and 805 volumes.