OPTIMALISASI DETEKSI WAJAH DLIB-HOG PADA CITRA INTENSITAS RENDAH DENGAN PREPROCESSING CLAHE

Abstract

Face decection using Dlib-HOG offers high performance under ideal lighting condition but significantly degrades when applied to low-light images. This study evaluates the effectiveness of Contrast Limited Adaptive Histogram Equalization (CLAHE) as a preprocessing method to enhance face detection accuracy under poor lighting conditions. CLAHE is applied to grayscale images to improve local contrast without introducing excessive artifacts, thereby making facial features more distinguishable for the HOG-based algorithm. Experiments were conducted on a facial image dataset with varied lighting conditions, comparing detection results before and after preprocessing. The results show a notable improvement in detection accuracy from 85.7% to 96.4% and a reduction in false negatives, with only a minimal increase in processing time. These findings confirm that CLAHE is an efficient and lightweight enhancement technique for improving the performance of Dlib-HOG on low-quality images.