Texture feature extraction and description using gabor wavelet in content-based medical image retrieval

Abstract

Gabor wavelet is one of the important methods for texture feature extraction and description in content-based medical image retrieval. Usually Gabor wavelet is used for a special scale set and a special direction set. However, this usually cannot extract the most discriminative texture features. In this paper, a new method for texture feature extraction and description is proposed. The method starts from whole scale space and whole direction space, and extracts time-frequency coefficients from each scale and each direction using Gabor wavelet. The energy is computed according to the coefficients, and dominant multi-scale and multi-direction fuzzy set is computed based on all energy computed. The standardized energy is used to measure the dominance of each element. Texture feature vector is computed according to the fuzzy set. The similarity measure is carried out between the fuzzy sets. When two images are of the same kind, their similarity measure is carried out between the texture feature vectors of two images. The experiments show that the method prosed in the paper has good retrieval performances for normal medical images.