<![CDATA[Diabetic retinopathy (DR) remains one of the most significant causes of preventable blindness worldwide, particularly in developing countries, where limited access to ophthalmological specialists hampers early detection. This situation underscores the urgency of developing automated and reliable systems capable of detecting key retinal abnormalities such as exudates, which serve as crucial indicators of DR progression. However, fundus images often suffer from challenges including uneven illumination, low contrast, and background noise, leading to reduced visibility of important features and lower accuracy in lesion detection. This study proposes a method for exudate detection in diabetic retinopathy fundus images that combines Color Dominance preprocessing and Gabor Filtering for feature extraction, followed by Support Vector Machine (SVM) classification. The preprocessing stage employs Contrast Limited Adaptive Histogram Equalization (CLAHE) within the Lab color space, guided by the variance of the blue channel to determine dominant color components and enhance local contrast without distorting natural color characteristics. The Gabor Filter, implemented at multiple frequencies (0.0471–0.3535 cycle/pixel) and orientations (0°–135°), extracts discriminative texture and frequency features from the enhanced images, which are subsequently classified using an SVM with a Radial Basis Function (RBF) kernel. Experimental results conducted on 900 fundus images from the APTOS 2019 Blindness Detection dataset demonstrate that the proposed method achieved an accuracy of 92.33%, sensitivity of 96.22%, specificity of 88.44%, precision of 89.28%, and an F1-score of 92.6%. The best results were obtained at orientations of 45° and 90°, corresponding to diagonal and vertical patterns, and frequencies between 0.0779–0.2135 cycle/pixel, which effectively captured the characteristic texture of exudates while minimizing background interference. The analysis revealed that the Color Dominance–Gabor–SVM pipeline not only improved feature visibility but also enhanced the discriminative capability of the classifier. This research contributes to the field of medical image processing by presenting a robust, interpretable, and computationally efficient approach for exudate detection, paving the way for the integration of computer-assisted diagnostic tools into early diabetic retinopathy screening programs.]]>
