This paper analyzes automatic estimating of emotional parameters from 2D photos based on the MPEG-4 rules. Each affective face image which obtained from a novel picture questionnaire is divided into three parts. The deformation function which calculates the displacements of vertices meshes within the influence of facial animation parameters are discussed. This paper develops geometric and morphological image processing methodologies to lifelike facial expression synthesis. Finally a 3D audio-visual avatar is created which gives the same expression corresponding to the input image. Experimental results are given to support our point of view. The experimental results show that our system properly applied in describing emotional facial expression.