Skin color detection by illumination estimation and normalization in shadow regions

Skin detection based on skin-color information has gained much attention because of its advantages in effective computation and rotation independent. An important challenge of skin color detection is how to detect skin color pixels located in shadow regions, which are corrupted by illuminant. In this paper, we propose an illuminant independent skin color detector. The detector consists of two steps: firstly it applies a novel color restoration method to enhance shadow regions; secondly, Bayesian skin color classifier is used to detect skin color pixels in color images. The proposed color restoration method applies Retinex algorithm which based on an advanced adaptive smoothing filter to normalize illumination. Firstly the input image is convolved with some adaptive smoothing mask windows, each mask weight in any window is computed via combining measures of the illumination discontinuity at each pixel; Then, Retinex theory is used to remove illumination estimation image from original image, and achieve the result image. The proposed skin-color detector present more efficiency and promising result when images under harsh illumination conditions. The experimental results prove that the new method can increase accurate hit of skin color pixel detection and maintain real-time performance.