Bifurcation Structures for Retinal Image Registration & Vessel Extraction

Exact extraction of the blood vessels of the retina is an important task in computer-aided diagnosis of retinopathy. This paper describes the development of an automatic image processing fundus and the analytical system to facilitate the diagnosis. The algorithm for the detection of the optical disc, blood vessels and exudates are investigated. The optical disc is identified by the method of the Sobel edge detector and LSR in the candidate area. Blood vessels and exudates are extracted by the method of Kirsch in different color components of the color image of the fundus. The processing results of the implemented methods are also presented. This paper also presents a new structural feature based registration function in the retinal image. The point coincident conventional methods depend largely branching angles single branching point. The feature correspondence through two images may not be unique because of the angle values like. In view of this, the record matching structure is favored. The branch structure comprises a branch point and its three neighbors master connected. The feature vector of each branch structure consists normalized angle and length of branching, that is invariant against translation, rotation, scaling, and distortion and even modest. This may considerably reduce the ill-posed nature of the pairing process, provided that the vascular pattern may be segmented. The simplicity and efficiency of the proposed method make it easy to apply alone or incorporated with other existing methods for formulating a hybrid or hierarchy scheme.