Pictures speak, but not directly to machines yet. There is a so-called semantic gap between the raw visual representation and the high level semantics incomprehensible to machines. Although proposed as a key area in artificial intelligence, image understanding lacks significant progress. An indication of the difficulty of generic image analysis is the domain-dependent nature of most image processing applications. A notable exception is however content-based image retrieval (CBIR), which, after more than a decade's research, has shed new light into image analysis and understanding.
We will present our approach in using projected feature maps for image collection profiling, and discuss how the visual data mining process working on a set of MPEG-7 feature descriptors may help to bridge the semantic gap.
Last modified: Thursday, 14-Oct-2004 14:48:02 NZDT
This page is maintained by Caroline Wills.