The Research On The Recognition Of Exudate In Fundus Image

In recent years, with the extensive use of electronic display devices, theincidence of ocular fundus diseases rate soared. Fundus images of fundus photographsproduced by fundus camera. Through the analysis of fundus images, can make adiagnosis and treatment of ocular fundus diseases. As an important application in thefield of Medicine Informatics, pathological changes of fundus image automaticrecognition for Department of Ophthalmology doctors diagnose disease provides amore convenient tool, development gets more and more quickly, have made certainprogress in the domestic and foreign. Exudate is an important disease characteristicsof ocular fundus diseases, to realize the automatic recognition of exudate fundusimages, is of great practical significance in clinical medicine.Identification of exudate fundus images in the main faced several problems asfollows: different light environment difference of brightness, contrast and otherobjective reasons, will cause a certain impact on image quality; different shootingselected angle and mode of operation can also cause the human impact; not the samelesions with age and different ethnic groups will exhibit different characteristics;exudate and other ocular structures such as optic disc tissue, macular, vascular, andsome diseases such as bleeding, hemangioma lesions are similar, increased thedifficulty of extraction and recognition.Exudate feature extraction and exudate classification are the two main researchcontents. First, through the use of morphological filtering on the fundus image regionof interest to be locked. Then the target extraction, comparison of5kinds ofthresholding methods, respectively is the Otsu method, iterative method, specifyingthe method, dynamic method and the method of threshold based on fixed ratio.Method for extracting fixed ratio using similarity based on optic disc and exudate,using disc occupation image area ratio as an approximation to the threshold, can betterachieve the target region extraction.Using extraction algorithm based on topological structure of outline contourtracking, and attributes of contour region extraction. The use of K nearest neighbor,naive Bayes method, backpropagation neural network to classify the target sample set,to achieve better classification results, at present on the disk, hard exudates, softexudative recognition sensitivity respectively83.3%,99.7%,88.7%.