Retinal image registration is an essential process that can significantly improve the accuracy of diagnosis and progression monitoring of retinal diseases. This technique involves comparing reference and test images using various image registration methods. The proposed approach in this paper is a specific feature region technique for retinal image registration that combines area and feature-based methods. This approach comprises five significant steps that ensure accurate registration. Initially, the vascular tree of the retina is extracted using an efficient Top-Hat operation with an optimal thresholding technique. This step elucidates the vascular structure of the retina, which is a crucial feature for registration. Next, the Harris-PIIFD detector detects the maximum point correspondence information and distinctive points based on the binary image (area-based). However, this process leads to redundant image-matching point correspondence, increasing the computation time. To overcome this, a redundant keypoints elimination (RKE) has been proposed to remove the redundant keypoints of the Harris-PIIFD algorithm, thereby reducing the overall computational load. Afterward, the features are matched with bilateral matching based on the best-bin-first algorithms for computing the similarity matrix for registering the images (feature-based). If the image pair is accepted, the simplest affine transformation modes control the points for the highest registration success rate. The simulation results on 134 pairs of FIRE datasets demonstrate the effectiveness and robustness of the proposed algorithm. This hybrid image registration approach is an efficient and reliable tool for retinal image registration, leading to more accurate diagnosis and monitoring of retinal diseases.
INDEX TERMSRetinal registration; retinal image; redundant keypoints elimination; bilateral matching.