INVESTIGATORS

Prof. Horace H S Ip, Mr. William Wong
 
 

This article presents a representation scheme that supports efficient indexing and retrieval of 3D head models. The proposed representation is based on the Extended Gaussian Images of 3D surface normals. When used together with a similarity measure, we are able to quantify visually similar 3D head models for retrieval purpose. Furthermore, by partitioning the 3D head model into distinctive facial regions, an hierarchical index for head model can be built. In our database, all the head models are represented up to about 3000 surfaces. We have developed a system for head model retrieval with an interface to allow users to define a visual query and to interact with the system. By comparing with Eigenfaces on the retrieval performance, our algorithm saves plenty of time on processing with similar results.
 
 

This work is part-sponsored by HKSAR RGC Research Grant No. CityU115/99E and CityU 1113/99E.
 
 

1. Horace HS Ip, Angus KY Cheng, William YF Wong, "Affine-invariant Sketch-based Retrieval of Images", CGI Conference 2001, City University of Hong Kong, pp.55-61, July, 2001.
2. William YF Wong, Angus KY Cheng, Horace HS Ip, "Concealment of damaged blocks by Neighborhood Regions Partitioned Matching", Proceedings ICIP 2001, Thessaloniki, Greece, Vol. 2, pp. 45-48, October 7-10, 2001.

Figures

Fig. 1 Head models retrieval system interface

Fig. 2 Hierarchical index structure for head models based on EGI