Different classification methods of fundus image segmentation using quincunx wavelet decomposition

Retinal vessel in fundus image is segmented with a comprehensive method for classification. The method is processed in four phases namely, preprocessing, segmentation, features extraction and classification, this method can be used on different images sets. Retinal vessels are enhanced by brightness preserving dynamic fuzzy histogram equalization (BPDFHE), separating enhanced image is used to detect the retinal diseases. Then these enhanced images are segmented by using quincunx wavelet decomposition for extracting features like first order statistics and gray level co-occurrence matrix (GLCM). The feature vector encodes information to handle the normal and abnormal retinal image and those features are classified using different classifiers (Adaboost, DSVM, ELMASR, EPLS, KNN, NB, NBFFS, OCPLS, RBFN, RF, SOWA, SVM and SVNN) and the performance is evaluated in detail. Blood vessel segmentation with this method is effective for retinal image computational analyses such as early retinal disease detection. Experimental results on three public retinal data sets like DRIVE, STARE and MESSIDOR and real time images are taken from Agarwal's Eye Hospital, Tirunelveli, demonstrating an excellent performance in comparison with retinal vessel segmentation methods reported recently.

Automated summary

The comprehensive method for retinal vessel segmentation and classification effectively detects early retinal diseases, extracts features and classifies them using multiple classifiers, showing superior performance compared to recent methods.