Stereo matching and 3-D reconstruction for optic disk images

3-D visualization of optic disk surface is very useful in diagnosis and observation of some eye diseases. It helps physicians in understanding and interpreting the stereo disc photographs (SDPs) which is widely used in clinical situations. This paper proposed a segment-based stereo matching algorithm, which represents the fundus structure as a Bayesian network and applies belief propagation(BP) to solve the maximum a posterior(MAX) estimation. Only ground control pixels(GCPs) of the BP results are retrieved and the dense disparity map is obtained by cubic interpolation and Gaussian blurring to ensure smoothness. The resulted 3-D retinal surface shows our approach is promising.