On the design and analysis of the software and data architecture of ROI-based image retrieval system

Abstract

Content-based Image Retrieval (CBIR) has been widely used in many areas such as crime prevention, the military, web searching, intellectual property, etc. The technology related to CBIR has been continuously developed, such as multi-dimensional indexing and features extraction algorithms. However, some issues still need to be resolved such as high-level semantics of images, human in the loop, etc. The application of CBIR in many professional areas is still under development. Thus further study and research are necessary and important to address those problems. In ophthalmology, physicians often need to compare retinal images in their collections with common properties for a specific purpose, or find similar Region of Interests (ROI) in different images, which is very useful as reference for current diagnostic. Although much effort in the pattern recognition has been explored and shown to be helpful, specific application for retinal images retrieval is still in its infancy. In this thesis, we propose a ROI (Region of Interests)-based image retrieval system, which supports finding similar ROI in the database based on the ROI content. After studying some CBIR techniques and reviewing several CBIR systems we proposed a Multi-layered 2D Matrix Indexing Scheme, which allows for retrieval of the similar ROIs from the database. The prototype architecture is presented and the data structure of the software system is discussed.