Research And Application On The Facial Feature Extraction Algorithm

Individual biological feature information has very strong uniqueness, and it’s also theimportant information to distinguish the different individuals and species. So thebiological feature information is always uses as the base of species clustering andindividual identification, and it is very important in the area of pattern recognition. Thefacial feature information is easy to observe, and it is also the important basis to identifythe different individuals. In all the human features, the facial feature information is themost direct and most easily presented in front of us, and human beings always distinguishthe different individuals through the facial feature information. So more and moreresearchers focus on the work about the facial feature information, and the applicationsrelated to the facial feature information are becoming more and more popular, facerecognition is one of representatives. The facial feature extraction as the most importanttechnology in the face recognition is directly related to location and matching to targetobject, therefore it has far-reaching significance to do the research to the facial featureextraction algorithm.This paper Bases on the facial feature, and deeply researches a classical statisticalmodeling methods-active shape model (ASM), and makes further analysis. In the processof reproducing this classic algorithm, the paper seriously resolves the main idea of thealgorithm and its inadequacies. At the same time of using the mainstream thinking ofalgorithm, the improved method is proposed to solve the inadequacies and further toimprove the robustness and efficiency of the algorithm. Specific research works are thefollowing:(1) The traditional active shape model (ASM), according to the principal componentanalysis, gets main components for the shape, and then the average shape model isobtained to approximately express target object model. However, when the differencebetween the feature shape model of target and the average shape model is large, it isdifficult to match. Therefore, this paper analyses the shortage of the traditional matchingalgorithm, and proposes an improved algorithm. This algorithm searches the most similarimage to the target object from the training sample set, and uses the shape model of themost similar image instead of the average shape model to approximately express targetobject model, so it can avoid the failure caused by the large difference between the featureshape model of the target and the average shape model.(2) The contour points distribution histogram (CPDH) under the polar coordinatesprojects the contour information to the gridding under the polar coordinates, and thehistogram will be created according to the number of feature points in each gridding, sothe histogram information directly reflect the contour information of the target object.This algorithm is not hard to understand, and the feature information presented in thehistogram are easily distinguished just with naked eyes, otherwise, this algorithm can keepinvariant to the translation and scaling on the target. CPDH has better retrieval accuracythan CSS and FD algorithm in the retrieval to the contour feature of shape. Therefore, this article will make the CPDH apply to the retrieval to face shape, and use CPDH to definethe face similar images to solve the problem of ASM model matching. In addition,according to the oval face shape, this paper also proposes deformed CPDH.