3D Reconstruction of the Retina Based on Un-Calibrated Fundus Images

With the aim of solving the problem that un-calibrated fundus images are unsuitable for the 3D reconstruction of the retina, a novel method is proposed based on the fusion of depth and disparity maps. To realize the stereo matching of un-calibrated fundus images, the images were adjusted by affine transformation matrix, and the retinal disparity map was solved using a semi-global block matching algorithm. To overcome the problem of missing absolute depth information from the disparity map, the retinal depth map was solved according to the optical system design principle of the fundus camera; and the optimal fusion of the two maps was performed according to the shortest Euclidean distance. Finally, 3D reconstruction of the retinal surface was achieved by OpenGL. The experimental results show that this method can preserve the relative and absolute depth information of the retinal surface, and that the 3D reconstruction effect is ideal. This method is robust and suitable for fundus images captured by most fundus cameras.