Automated control point detection, registration, and fusion of fuzzy retinal vasculature images

Multi-modality biomedical imagespsila feature detection, registration, and fusion are usually scene dependent which requires intensive computational effort. A novel automated approach of the multi-modality retinal image control point detection, registration, and fusion is proposed in this paper. The new algorithm is reliable and time efficient, which implements automatic adaptation from frame to frame with a few tunable thresholds. The reference and input images are from two different modalities, i.e., the angiogram grayscale and fundus true color images. Retinal imagepsilas properties determine the fuzzy vessel boundaries and bifurcations. The retinal vasculature is extracted using canny edge detector and the control points are detected at the fuzzy vasculature bifurcations using the adaptive exploratory algorithm. Shape similarity criteria are employed to match the control point pairs. The proposed heuristic optimization algorithm adjusts the control points at the sub-pixel level in order to maximize the objective function mutual-pixel-count (MPC). The iteration stops either when
f

_MPC
reaches the maximal, or when the maximum allowable loop count is reached. The comparative analysis with other existing approaches has shown the advantages of the new algorithm in terms of novelty, efficiency, and accuracy.