A high resolution grammatical model for face representation and sketching.pdf

A High Resolution Grammatical Model for Face Representation and Sketching Zijian Xu, Hong Chen and Song-Chun Zhu Departments of Statistics and Computer Science University of California, Los Angeles {zjxu, hchen, sczhu}@stat.ucla.edu Abstract In this paper we present a generative, high resolution face representation which extends the well-known active ap- pearance model (AAM)[5, 6, 7] with two additional layers. (i) One layer refines the global AAM (PCA) model with a dictionary of learned face components to account for the shape and intensity variabilities of eyes, eyebrows, nose and mouth. (ii) The other layer divides the face skin into 9 zones with a learned dictionary of sketch primitives to represent skin marks and wrinkles. This model is no longer of fixed dimensions and is flexible for it can select the diverse rep- resentations in the dictionaries of face components and skin features depending on the complexity of the face. The selec- tion is modulated by the grammatical rules through hidden ”switch” variables. Our comparison experiments demon- strate that this model can achieve nearly lossless coding of face at high resolution (256 × 256 pixels) with low bits. We also show that the generative model can easily gener- ate cartoon sketches by changing the rendering dictionary. Our face model is aimed at a number of applications includ- ing cartoon sketch in non-photorealistic rendering, super- resolution in image processing, and low bit face communi- cation in wireless platforms. 1. Introduction Human faces have been extensively studied in vision and graphics for a wide range of tasks from detection[18, 16], classification[19, 10], tracking[6], expression[13], animation[11, 15], to non-photorealistic rendering (portrait and sketch)[3, 4], with both discriminative[17, 10, 3] and generative models[9, 6, 11] developed in the literature. The selection of a representation and model depends on two fac- tors: (i) the objectives of the task and its precision request, an