Combing Adaptive Pulse Coupled Neural Network and Maximal Categories Variance Criterion for Blood Vessels Automatic Detection in Fundus Image

A new blood vessels automatic detection method in fundus image combing adaptive Pulse Coupled Neural Network (PCNN) and maximal categories variance criterion is proposed. In preprocessing, Contrast Limited Adaptive Histogram Equalization (CLAHE) and two-dimensional Gaussian matched filtering are adopted to improve the contrast between blood vessels and background. Then based on simplified PCNN model and maximal categories variance criterion, the preprocessed fundus image is segmented. In image processing, the linking strength of each PCNN neuron is usually a constant. In order to overcome the limitation, pixels Energy Of Laplace (EOL) is chosen as the linking strength of corresponding PCNN neuron, thus PCNN can adjust its linking strengths according to pixel features adaptively. Finally, the final blood vessels detection result is obtained via postprocessing including area filtering and breakpoint connection. The experiments implemented on the Hoover fundus image database show that the method has relatively higher robustness, effectiveness and reliability.