Detection, Based On The Brightness Of The Color

At present research on skin color detection is a hot field in image recognition, and it is also pre-process of skin color based on some application such as face detection, hand gesture recognition, porn image recognition and filtering and so on. A crucial processing step for the success of such computer vision-based systems is robust detection of skin color. Though a lot of skin detection techniques appear, they still impracticable in most applications, the skin color detection technique under complicated illumination and background is an open issue. Therefore, the thesis lays particular emphasis on studying the skin color detection technique under the complicated illumination.This thesis briefly introduces the elementary knowledge of skin detection at first, and then research on some key techniques is done, finally we have finished the following main task:1. Some kinds of color space usually used are analyzed. A method that is combined with HSV color space and YCbCr color space is adopted to detect skin color here, and it can eliminate the interference from light. Both color spaces are light-chrome separated, and YCbCr is used to ascertain the light value of pixel since the simplicity of computation of Y on light behalf, and another is used to divide skin color space since its good cluster effect on H-S.2. Some different skin color models are analyzed and a conclusion is drawn that a kind of model with Bayes based on HSV color space is best. For get best skin color detection model, three models adapted in large specimens are compared: threshold, histogram based on HSV(normalized lookup table and Bayesian classifier included), the outcome of test demonstrate the correctness of conclusion.3. Three detection models with different light are set up to handle changing light conditions respectively. To solve the light problem in skin color detection, that is, to solve problem about high-or-low light existing possibly, three detection models that include high light, normal light and low light are constructed.4. A method that image will be divided into several blocks to avoid higher or lower light locally is adopted. After blocking, every block will be processed by its average light value.