Texture Characteristic Extraction for Dominant Directions in Content-based Medical Image Retrieval

The paper presents a texture characteristic extraction method for dominant directions. Enhanced Fourier spectrum is computed for an image in a medical image database. Direction histogram is computed from the enhanced Fourier spectrum. The set of dominant directions is computed according to the histogram. Gray co-occurrence matrix extracts texture characteristics based on the set. Euclidean distance is used to measure similarity between the images. Experiments show that the method in the paper show good performance for directional texture images and the same performance as currently widely used methods for non-directional texture images.