Nearest Target for Retina Images Based On Canny Edge Detector and Ensemble Classifiers
For retina problems diagnosing, valuable information to ophthalmologists are provided by Retinal fundus images. Curing rate can be enhanced using early detection and it may prevent blindness. Using retinal fundus images, medical experts diagnosis retinal problems like retinitis pigmentosa and diabetic retinopathy. Mutual Information (MI) optimization is initialized as coarse localization process, where optimization domain is narrowed and local optima are avoided in recent works. In addition, Improved Support Vector Machine (ISVM) technique is used for performing retina image’s nearest template and it is used with registration based on area, provides a robust technique. However, proper detection of retina image’s edges are not done using these techniques and concerned region’s visibility and perceptibility. Performing Retina image’s nearest template according to single classifiers leads to degradation in classifier accuracy. So, for avoiding this issue, a four stage framework is proposed in this work. Fuzzy clustering algorithm based noise removal is done in the first phase, where similarity among un-noisy and noisy pixels are computed. Canny edge detection operator based edge detection and enhancement are done at the second stage. Dimension reduction is done in the third stage, where Mutual Information (MI) optimization is initialized as coarse localization process, where optimization domain is narrowed and local optima are avoided. Fuzzy neural network (FNN), Probabilistic neural network (PNN), Adaptive Neuro Fuzzy Inference System (ANFIS) classifier’s ensemble is used for computing retina image’s nearest template in fourth stage. For STARE dataset, with respect to run time rate, success rate and mean error rate, proposed model’s robustness is demonstrated using experimental results.