Face Detection Based On Template Matching

Face detection is the study focus and difficult point in computer vision and computer figuring..At present, there are two kinds of face recognition and detection methods according to complexion characteristic and grey degree characteristic. According to the method of complexion characteristic, the first is used to make certain complexion model, then proceed the skin color partition, according to their similarity in colour and the relativity on the space out possible district, make use of the district geometrical characteristic or ash degree characteristic to verify if it is a face of person at the same time. The common face characteristic detection technique basically come down to three types,namely: method based on geometrical characteristic, method based on template and method based on model. The method based on geometrical characteristic is represented by Snakes method and Deformable Template method.The method by based on template include the characteristic face method,the method according to related matchjinear analytical method,singular value decomposition method,neural network method, dynamic connection match method etc The most representative and successful method is the face characteristic method. The method based on model includes Active Shape Model ,Active Appearance Model, Direct Appearance Model ,etc.This paper,first, proposes the calculation method which is based on model face detection on the foundation of summarization on the relevant documents about face detection and recognition technology .This method uses the average face template match calculation directly. As the eyes play a key part in person's face measuring and recognition, the paper divide the template into the eyes template and facetemplate . The template is structured through taking the average on a lot of personal face samples.First ,I drew out the face district according to the sample diagrams that we took by hands as the face sample,Then, proceeded the dimensions standardization and grey distributing standardization to every face sample. After that ,we took grey average on these sample diagrams to got the average face picture. Further re-sampling to certain size, we got the primitive template, copying the eyes part of the primitive template and standardized grey distribution. We got the double eyes template.Carrying on grey distribution standardizing primitive template ,we got the person's face template.In order to adapt to different shapeas much as possible, we can pull the face template by proportion of length and width to constitute the person face template. We first took rough selection with eyes template while matching,then used person's face template to further match to determine the proper person's face area in the picture that were waiting for dectection, then use gradient rule of the gradient mosaic images to get rid of the part of false face.Finally, we send the pictures district that passed on detection just to the sub-space of person's face based on singular value characteristic distribution to verify,so we can confirm if it was person's face.If it is, we can mark out the position of the person's face.