A novel color region homogenization and its application in improving skin detection accuracy

Abstract

Pixel based color classification is a primary process in some advanced intelligent user interfaces. It is especially applied for the human skin detection purpose and consequently in the more effective interactions through the face or hands gesture. The extracted areas by the skin classifier are usually scattered and not homogenous based on the nature of the images and the process. It is while in some applications like facial perception we will need more accurate and smoother skin region extraction.There are few methods applied in some previous skin detection researches but there is no special evaluation or judgment particularly on this topic. Here in this work it is tried to introduce a simple and effective method for region homogenization after a pixel based color classification based on distances in both color and spatial domains. Its higher accuracy is shown in compare of three other methods and then discussed about.