Métodos de pré-processamento para a correção da iluminação irregular em imagens de fundo do olho

Abstract

This work compares some preprocessing methods that perform contrast enhancement on color eye fundus images. Retinal images usually contain illumination artifacts that are occasioned during the acquisition process. One way to get a better contrast and sharpness is the preprocessing methods application. Also, these methods reduce the noise present in eye fundus images. Preprocessing methods aim to enhance structures optic disc and injuries exsudatos on the images. After methods application, using comparision strategies we determine the best preprocessing algorith for an image group with similar characteristics. Three preprocessing methods were chosen: FAS (Sequential Alterante Filters), CLAHE (Contrast Limited Adaptive Histogram Equalization) and the Morphological Filter denominated TOPHAT. The experimental results based on 30 imagens from DIARETDB1 eye fundus images database. The results analysis and comparison aim to find the most effective preprocessing method to eliminate unwanted artifacts and lighting errors in retinal images. To validate the preprocessing methods we used two metrics: sensitivity and specificity. Our experimental results demonstrated that the best preprocessing method in face of the images characterists variety is TOPHAT. The best execution with TOPHAT used structuring element on 40 with 84% of sensitivity and 74% of specificity