Searching digital biomedical images is a challenging problem. Prevalent retrieval techniques involve human-supplied text annotations to describe image contents. Biomedical images, such as pathology slides, usually have higher resolution than genral-purpose pictures, making it additionally difficult to index. Precise object segmentation is also extremely difficult and is still an open problem. We are developing a multiresolution region-based retrieval system for high-resolution biomedical image databases. The system, based on wavelets, the IRM (Integrated Region Matchign) distance, and image classification, is highly robust to inaccurate image segmentation and various visual alterations. Tested on a database of more than 70,000 pathology image fragments, the system has demonstrated high accuracy and fast speed.
展开▼