Wavelet-Based Computer-Aided Detection of Bright Lesions in Retinal Fundus Images

摘要

Computer-aided detection and diagnosis of diabetic retinopa-thy with retinal fundus images is the necessary step for the implementation of a large scale screening effort in regions where ophthalmologists are not available. In this paper we propose computer-aided binary detector of bright lesions in retinal fundus images. It is based on wavelets for mul-tiresolution feature discrimination and support vector machine (SVM) for classification. After thresholding the sub-band images resulting from the Isotropic Undecimated Wavelet Transform (IUWT) decomposition of the input image, we employ an approach based on the image Hessian eigenvalues and multi-scale image analysis, for designing good feature descriptors of bright lesions. These are afterwards used in the SVM model classifier. Experimental results on our current data set show that the proposed method is efficient and achieves a very good success rate.