2D Facial Transformation Based On Maximum A Posteriori And Subspace Learning

Human face plays an important role in social communication as media to directly convey information, so in the era full of information it is utmost important of processing face images automatically by computers for wide applications. However, the complicated geometric features of human face makes learning a 2-dimensional face transformation require large scale of training images and high level of computation complexity. Hence based on the actuality as above, in this paper we propose a method based on Maximum A Posteriori criterion and subspace learning to learn 2-dimensional face transformation by using small scale of training set, and solve human face aging and colorization of gray-scale facial image by the novel approach.For learning the process of 2-dimensional facial age transformation, an ideal training dataset should contain series of the faces from childhood to old age corresponding to each identity. However, such a training set is difficult to obtain in practice. Hence a novel method based on Maximum A Posteriori is proposed and the aging face is simulated automatically through a two-step process:global geometric affme transformation and local texture synthesis by sparse representation.And for the 2D facial image color transformation, especially recovering the true color for gray image, a new prediction technique is devised which is launched by Maximum A Posteriori criterion to evaluate the color. The core of correctly predicting the color is to minimize the global energy function based on a patch-based system and learn a feature subspace with locality.The experiments of the two 2D face transformation applications show the newly proposed method in this paper outperforms the existing techniques.