Knowledge Extraction and Refinement From Multi-Feature Images Through (Re-)Clustering

摘要

This paper presents a generic knowledge-guided seg-mentation system that is capable of segmenting multi-feature images then extracting and refining knowledge from an image. It provides a framework for processing a series of images of the fixed location. Unsupervised fuzzy clustering and basic image processing techniques can be effectively incorporated into the system utiliz-ing the guidance of a knowledge base. A major portion of the knowledge comes from images ground-truthed by domain experts. These ground-truthed images are seg-mented by the system to extract cluster labeling rules trough (re-)clustering using a fuzzy clustering algo-rithm. The cluster labeling rules are stored in the system konwledge base and later refined whenever new training images are available.