Co-occurence trees: a dynamic solution for texture feature extraction

Texture feature extraction has shown valuable results in medical image analysis, segmentation and classification. Thus, efficient methods for storing the textural information and computing the textural features of an image must be devised. Co-occurrence matrices are powerful for discriminating different textures but are not efficient in terms of memory requirements. Due to their dependency on the number of gray levels in the entire image a large amount of space is being wasted while time complexity for the feature extraction operations is raised. This paper presents a novel, dynamic approach to organizing the textural information which provides efficiency in both storage requirements and computational time, being dependent only on the number of gray levels in the examined local region. Furthermore, an updating algorithm which achieves a near half reduction in time is presented and evaluated. Finally, results are presented and discussed which clearly indicate the superiority of the proposed approach.