Texture Features for Browsing and Retrieval of Image Data
IEEE Transactions on Pattern Analysis and Machine Intelligence
Hi-index | 0.00 |
Differentiation of pathological architectural distortions (ADs) in mammograms is investigated with the aid of numerical description of local texture orientation. Challenging purpose of our long-term research is effective method of ADs detection in mammography. It is reasoned by significant limitations in efficiency of commercial and research detectors of pathological ADs. Extracted and enhanced texture structures are analysed in order to automatically indicate the small regions with potential pathologies. Gabor filtering is proposed as promising extractor of texture orientation according to our previous studies and the reported results in literature. However, adjusting Gabor filters (GF) to extremely changing manifestation of ADs spicules, diversified in size, forms, shapes and intensity is still open question. Therefore, we optimize GF by impulse response shaping, adjusting of angular resolution and taking into account local image activity. Finally, a probability map of ADs appearance is estimated using the results of Gabor filtering of mammograms. Experimental verification of Gabor maps efficiency leads to 82% sensitivity at 2.64 false positives per test case (FPR) if we use only Gabor filtering to detect ADs or 88% sensitivity at FPR = 4.4 in case of additionally ADs recognition stage on selected ROIs by texture orientation enhancement, analysis and extraction.