Key point localization for 3d model generation from facial illustrations using SURF and color features

Risako Aoki, Shun Aoki, Yakumo Ohtagaki, Ryusuke Miyamoto

Abstract: This paper proposes a novel scheme to improve localization accuracy of facial key points in facial illustrations. The proposed scheme estimates the location of facial key points considering global structure of facial key points evaluated by RANSAC like scheme where local evaluation is performed with SURF and color features. Experimental results using a data set composed of facial illustrations show that the estimation error can be reduced to about 8.93 pixels per a key point.