Based skin human detection in natural and complex scenes

In this paper, the authors presented a new method to detect human skin in complex scenes, without priori information, based on skin-color approach. The work is performed in four stages. First offline skin-color model is developed using testing images. Second the skin-pixels in the image are detected. Third a merging procedure based on variance analysis, is applied to form skin regions. Fourth online skin-color model adaptation is applied to update the model parameters. The authors introduced variance measure in the homogeneity analysis. The maximum likelihood estimation for online adaptation was proposed. The system is tested on 50 images (100 persons); the results and performance are comparable to the reported works. The conclusion and future work are suggested.