Interactive retrieval in facial image database using Self-Organizing Maps
Zhirong Yang and Jorma Laaksonen
In: IAPR Conference on Machine Vision Applications, 16-18 May 2005, Tsukuba Science City, Japan.
Content-based image retrieval in facial image collections is required in numerous applications. An interactive facial image retrieval method based on Self-Organizing Maps (SOM) is presented in this paper, in which multiple features are involved in the queries simultaneously. In addition, the retrieval performance is improved not only within queries for current user but also between queries by long-term learning from other users' relevance feedback. In that way recorded human intelligence is integrated to the system as a statistical feature. The work constituting this paper has been incorporated into our image retrieval system named PicSOM. The results of evaluation experiments show that the query performance can be substantially increased by using multiple features and the long-term learning.