Full-Automatic Optic Disc Boundary Extraction Based on Modified Local Binary Fitting Model in Retinal Image

Accurate optic disc (OD) segmentation plays a significant role for retinal disease diagnosis. In this paper, a full-automatic segmentation approach named double boundary extraction for the OD segmentation is proposed. It is consisted of two main stages: first, the satisfying initial level set contour for OD boundary is extracted by an adaptive method based on unsupervised learning technology and statistical method; second, the proposed LSO model combined with prior information of OD can effectively obtain the final OD boundary. Extensive experiments on the publicly available DIARETDB 1 database demonstrate that the proposed method has the advantage over the state-of-the-art methods.