Research On Automatic Detection Methods Of Exudates On Diabetic Retinal Images

Diabetic retinopathy is a serious complication of diabetes,it is an important factor that causes vision loss and even blindness in adult diabetics.Hard exudates is one of the early features of diabetic retinopathy,the automatic detection technology of hard exudation has an important practical research value.In this paper,we have carried on the thorough research to the automatic detection methods of the hard exudates of the diabetic retinopathy.The main research contents of this paper are as follows:(1)Introduction to the basic theories and methods of hard exudation detection.Methods of image preprocessing include Gamma correction,CLAHE algorithm and enhance image in YIQ space.They can well solve the problem of low contrast and uneven illumination.The theories for extracting hard exudates candidate include morphological reconstruction techniques and Otsu segmentation algorithms.Optic disc detection methods include visual saliency theory,Hough transform and graph cut theory.(2)Proposed a hard exudates candidates extract algorithm based on morphological reconstruction techniques and Otsu algorithm.We enhanced retinal images contrast by Gamma correction,CLAHE and enhance image in YIQ space.Then,the hard exudates candidates were obtained by morphological reconstruction technique and the segmentation method of Otsu.In order to avoid missing the small hard exudates,we overlay the detection results of different brightness images.Finally,we post-processing the hard exudates candidates by using the method of Hough transform and graph cut and Itti algorithm,the optic disc is segmented and removed in the hard candidates.In this way,we can improve the accuracy of hard exudates candidates extraction.(3)To achieve precise detection of hard exudates in diabetic retinopathy images.In order to train SVM classifier and Ensemble classifier,we extract color features,brightness features,Gaussian filter average response,Laws texture features and Gabor features of the hard exudates candidates.Using these classifiers we can differentiate between hard exudates and other things,eventually we get precise hard exudates.We classify hard exudates candidates with extracted texture features and non-extracted texture features respectively by using SVM classifier and ensemble classifier,evaluate the classification results by sensitivity and mean predict value.Experimental results indicate that the Laws texture and the Gabor filter feature have significant effects on the classification of hard exudates.Compared with other hard exudates detection methods,the proposed method achieves high sensitivity and average predict value.The method proposed in this paper has some advantages and it is a very effective method for detecting hard exudates.