Some Strategies To Improve The Objict Profile-Based Image Retrieval Model And Its Prototype

An image retrieval system can be objecthapeased. Within such a system, objects are usually extracted from images with the outlines drawn by hands. Although the elemental but hard problem of automatic image segmentation can be thus avoided, the practicality of system is also weakened. In this thesis, four strategies are proposed to improve the existing objecthapeased image retrieval system. They are: semiautomatic recognition of objects with the help of shapeased segmentation and manual interference in term of object profiles; allowing the regions merging be postponed to the stage of user querying; flexible implementation schemes for regions merging: and detection and separation of texture regions from the image before segmentation. A prototype of objectrofileased image retrieval is then built upon these strategies as the base and framework, and some primary tests are also made with this system.