A Region-Analysis Subsystem For Interactive Scene Analysis

Abstract

An Interactive Scene Interpretation Syste• (ISIS) is being developed by Stanford Research Institute's Artificial Intelligence Center as a tool for constructing and experimenting with manmachine and automatic scene analysis methods tailored for particular image domains. A region analysis subsystem was developed recently based on the work of Brice and Fennema, and Yakiaovsky. Using this subsystem a series of experiments was c~nducted to determine good criteria for initially partitioning a scene into atomic regions and merging these regions into a final partition of the scene along object boundaries. Semantic (proble.dependent) knowledge is essential for complete, correct partitions of complex real-world scenes. An interactive approach to semantic scene segaentation was developed and demonstrated on both landscape and indoor scenes. This approach provides a reasonable methodology for segmenting s99nes that cannot be processed completely automatically at present and is a promising basis for a future fully automatic system.