An Adaptive Thresholding Method for Facial Skin Detection in HSV Color Space

Skin detection plays a crucial role in facial beautification. Traditional skin detection methods mostly use fixed thresholds within color spaces, leading to low accuracy and unsatisfactory results in facial beautification. This article proposes an adaptive thresholding method for facial skin detection in HSV color space. The method obtains the region of interest through facial feature points and then adjusts the mean of hue, saturation, and value channels in the region of interest by adding and subtracting multiples of their respective standard deviations. These adjustments are utilized to define a new threshold for skin detection and facilitate the completion of the skin detection process. The experimental results on the CelebA-HQ-img_Face+ and Helen_Face+ datasets indicate that, compared with traditional fixed threshold detection methods, multicolor space threshold skin detection method, and traditional adaptive methods, our method performs the best in pixel accuracy (PA) and intersection over union. From the results of the beautification images, our method achieves more accurate skin localization and provides effective support for facial beautification.