Improved Search Strategy for Active Shape Model of Face Alignment

Abstract

Face alignment is to look for the location of face features such as eyes, nose and mouth in the face image. Based on the face alignment of the active shape model, the point distribution statistical probability model was established, and the mahalanobis distance between the local features was calculated based on the iterative optimization of the gray information around the feature points, so as to find the position of the feature points in the new image. The traditional search process is that after initializing the shape, affine transformation parameters are calculated according to the updated position of the feature points, until the parameters converge, that is, the face alignment is completed and the position of the feature points is obtained. However, this process requires a lot of calculation, and the weights in the updating process are obtained by experience. In this paper, a search process based on genetic algorithm is proposed. The initial parameter of affine transformation is the value when training the model, and the objective function is the new shape vector multiplied by affine transformation parameter to get the difference between the result and the average shape, and the search is completed when the difference converges or reaches the number of evolution.