A probabilistic detection-based approach to skin and freckle segmentation.

Accurate freckle segmentation is essential for dermatological assessments and cosmetic applications, but existing lesion detection techniques are primarily designed for well-defined skin abnormalities such as melanomas and tumors, making them less effective at capturing subtle features like freckles. In this study, we present an automated freckle segmentation framework that integrates the Gaussian Mixture Model (GMM) and the Viola-Jones algorithm for skin segmentation, coupled with an energy map-based approach for freckle detection. The process begins with image is clustered using GMM, followed by facial region detection with the Viola-Jones algorithms. A post-processing step then segments the selection of the skin region. Subsequently, an energy map is generated by combining the blue and saturation channels, while Contrast-Limited Adaptive Histogram Equalization (CLAHE) and morphological operations enhance freckle contrast. The final segmentation is achieved through binarization and additional post-processing techniques. Quantitative evaluations demonstrate that the proposed method surpasses conventional approaches in recall, Intersection over Union (IoU), and Dice coefficient, highlighting its effectiveness in accurate freckle detection and segmentation. These findings indicate that, with further refinement, the proposed framework holds significant potential for applications in both clinical dermatology and cosmetic science.