Performance Analysis in Common Color Spaces of 2D Gaussian Color Model for Skin Segmentation

The aim of this paper is to determine the most effective color space through common color systems for 2D (two dimension) Gaussian Color Model to segment human skin in image. Skin segmentation generally is preprocessing step in many algorithms such as face detection algorithm, hand detection algorithm, etc. RGB, HSV, YIQ, and YUV can be categorized as common color spaces. Choosing appropriate color space is very important to increase algorithm efficiency. To this end, 2D Gaussian Color Model is built in each common color system using face photos of 50 European people, 50 Asian people and 50 African people from internet randomly in order to detect skin regions. 150 color face skin images size of 50x50 are utilized to get 2D Gaussian Color Model parameters. ROC (receiver operating characteristics) curves are used to performance analysis of the mentioned model in stated color spaces. Experimental results demonstrate that the most exact skin segmentation is obtained in color space for 2D Gaussian Color Model.