The Study Of Facial Feature Location Based On Active Shape Model

The facial feature localization is that computers locate the position of variouscharacteristics of the face image automatically. The facial features localization playsa very important role in the field of image processing, it provides data for theresearch of face recognition, expression analysis and face-tracking, and it will be agreat help in terms of target tracking and facial animation. Face is non-rigid and hascomplex three-dimensional structure. Appendage occlusion, illumination, pose andthe differences of camera equipment will causes great distress of the computer locatefacial features, so can not be accurately positioned features points.Active Shape Model (ASM) is a popular method of facial feature localization inrecent years. Active shape model can get the statistics of the image, through thetraining of the model, and can restrict the image within a reasonable range. Theactive shape model is divided into two parts, the training phase of the model and thesearch stage. In the training phase of the model, it can get the global shape model ofthe representation image global information and the local model of therepresentation feature point local information by manual calibration the location offeature points coordinates; in the search stage, firstly use the local model to find thelocation of local features in the image; after the search of feature point is over, usethe global shape model to restrain until the positioning of the feature points areaccurate or the number of iterations reaches a certain, the search ends. This papermainly improves the active shape model, and the improved methods can locate thefeature points more accurately. The main content of this paper is as follows:1. Facial feature location based on feature fusion. According to active shapemodel for facial feature point location is sensitive to the initial model and the localmodel only based on the gray level information is too single, this paper proposes animproved active shape model based on feature fusion. This algorithm combineswavelet transform and SUSAN operator to extract the corner point on a human's faceas an initial position of facial feature point. The fusion method of local binarypattern and the original local modeling method are used to establish the local model,which can express the details of feature point more completely.2. Facial feature location based on multi-population genetic algorithm. Thispaper proposes a new multi-population genetic algorithm base on depth study ofmulti-population genetic algorithm. This paper applies this new method to active shape model, and uses the genetic manipulation of chromosomes in the geneticalgorithm to replace the search process of the original method.