An Iconic and Semantic Content Based Retrieval System for Histological Images

This paper describes an intelligent image retrieval system based on iconic and semantic content of histological images. The system first divides an image into a set of subimages. Then the iconic features are derived from primitive features of color histogram, texture and second order statistics of the subimages. These features are then passed to a high level semantic reasoning engine, which generates hypotheses and requests a number of specific fine feature detectors for verification. After iterating a certain number of cycles, a final histological label map is decided for the submitted image. The system may then retrieve images based on either iconic or semantic content. Annotation is also generated for each image processed.