A Comparison of Color Models for Color Face Segmentation

Face localization is the first step in several applications such as face tracking, person identification, expression recog- nition and eye tracking. Face localization can be performed by segmentation using the color of the skin. Color images can be represented in several color models. This work presents a performance comparison between several color models including RGB, HSI, CIELab and YCbCr. The best performance in terms of classification error is achieved by the HSI and YCbCr models. However, due to the fact that all images are capture in the RGB color model, a transformation be- tween this model and the other models must be performed. The best performance in terms of execution time is achieved by the YCbCr model. An important speed-up can be achieved by downscaling the original image. Results show that, in the case of the HSI model, a downscale factor can speed up the process up to a 28% while a factor of 4 can speed up the process as much as 68%.