Algorithms Research And Implementation For Face Detection System

In recent years, the medicine and cosmetology professions have obtained vigorous development, the skin's health care and cosmetology more and more receive people's attention. Now the intelligent diagnosis system which can test the skin's conditons has appeared. But it needs manual gathering different skin regions to examine different skin targets. In order to realize more automated skin diagnosis, this paper introduces the on-line automatic diagnosis system which only needs a colored single positive forehead image, according to this image, it can automatically diagnose the skin's conditions of different regions, but does not need multiple manual image gathering and diagnosis, so we can say this system has realized truly automatic and comprehensive diagnosis for the skin. To achieve this kind of system: first, it must pick up the face area and facial organs from this image automatically; then divide the face area into different regions and diagnose different regions for different skin targets, this system finally achieves overall diagnosis for the skin targets such as skin color, speckle, wrinkle, moisture and oil. This paper is mainly responsible to study and realize the first step of this system-Picking up the face area and facial organs automatically. The specific contents include:1)Using the characteristic of skin color to detect face area. The detection is realized through the process of establishing Gauss Model for face color-threshold segmentation-Mathematics Morphology proposal. After face area detection, in view of this system's characteristic and following request, add the module of ear removing and finally confirm the rectangle area of face.2)Picking up the facial organs in the face rectangle area. Eyes location is mainly refered to the complexity algorithm and improve on it, at the end of the algorithm, pick up the first four points which have the biggest structure central degree as the suspicious centers of eyes, then through adding some judging terms, extract the real two centers of eyes successfully at different conditions, this method well improves the eye detection rate; For mouth detection, it's based on the lip's characteristics of bigger difference between R and G components and the stable distribution range on Q component; For nostril extraction, mainly use the low-gray characteristic of nostril, realize the nostril detection through picking up the lowest five percent pixels in gray of the nostril's searching area.3)In foundation of two previous steps, detecting the two points of intersection between neck and cheeks according to the cheeks and neck's position relations and their respective shape features. Then determine the concrete position of chin according to the face's distributed rule. Finally, using conic section to approximate the chin outline and remove the neck area. In the end, confirm the accurate face edge and complete the pick-up of face area and facial organs.Test all the algorithms using two hundred images gained through digital camera. The experimental results indicate that the algorithms of face detection and facial organs extraction which this paper propose all possess high detection speed and detection rate.