Registration of multimodal and temporal images of the retina using a combined feature-based and statistics-based method

Registration of retinal images is important for doctors to diagnose because it merges information of several images. This paper presents an automated multimodal and temporal retinal image registration method which is especially robust to infrared radiation (IR) fundus images, which may suffer terrible imaging artifacts, such as extremely low contrast, overexposed edges, or invisibility of blood vessels. The method we propose combines the feature-based registration method and the statistics-based registration method by using the OD detection method based on Line operator to detect the OD for the fundamental registration and use the maximal mutual information method for fine registration and improve them by down-sampling, getting Region of Interest (ROI), bilateral filter and so on to get a better effect. Experiments over optical coherence tomography (OCT) and IR dataset show that our proposed method outperforms other methods by obtaining an accuracy of 82% and an average time consuming of 5.1 second.