Study of Representing Facial Parts i n an Automatic System for Generating Line Drawing Based Portraits

Abstract

This paper explores conditions for accurate generation of portraits and how and whether noses should be represented in the portraits. ASM is applied to the original face image so that feature points in the face are detected. Using training portraits drawn by illustrators and their original faces, neural networks are trained. An unknown person’ face’s feature points are inputted to the trained neural networks so that that person’s portrait is automatically generated. As a result of experiments, multiple neural networks assigned to facial components tend to give smaller errors between the generated and manually drawn portraits than a single neural network assigned to the entire face. And the result of the compare test between portraits with nose and portraits without nose show that it is necessary to generate nose in portraits.