Bridge over the Troubled Semantic Gap

BABAGUCHI Noboru

The ultimate goal of media understanding researches is to acquire semantics from images and videos. However, the semantic gap, which means the lack of coincidence between the information that can be extracted from the visual media and the interpretation that humans give for it, makes media understanding very difficult. Since the gap between low level image/video features and high level concept descriptions is very large, we have not built a bridge over the troubled gap until now. In this lecture, we discuss three different approaches to bridge the semantic gap, tracing back to our research activities. The first is statistical pattern recognition or machine learning with extensive use of large-scale corpus with ground truth. This has been one of the most successful approaches in recent years. The second is information integration or multimodal analysis. In this approach, it is of great importance to view the real world in a multi-modal fashion as well as to make use of the Web space as a large-scale knowledge source. The third is human-centered computing; in other words, human-in-the-loop systems. Because such a system is capable of incrementally learning the correspondence between high and low level descriptions through human-system interaction, its performance of understanding media semantics can be increased.

Bridge over the Troubled Semantic Gap

Bibliographic Information

Search this article

Description

Journal

Keywords

Details 詳細情報について

Export

Report a problem