| In the 21st century, the digital technology develops at a very fast speed and more and more digital images are flooding our life. How to search for and locate the images we need accurately and quickly, which is a problem we should solve immediately, is the main aim of the content-based image retrieval (CBIR). From image database, it retrievals the images that are similar to the sample image that a user inputs by making best use of the inherent features (such as color, texture, shape and structure, etc.) of image itself.However, the current research of CBIR in the world is all concentrated on the extraction and rendering of the low level features of the image and relevance feedback, while at the meantime lacking the general description of the image. This undermines the wide use of feature description and retrieval system, and hence causes the waste of resources.This paper analyses the image retrieval systems of the world firstly and points out their advantages and drawbacks. Then it introduces the theory and traits of CBIR, and conducts a wide and deep research on the retrieval and rendering of image color. After that it introduces the basic knowledge of MPEG-7 and the methods of image similarity matching. Finally, by making use of the extraction methods and relevant knowledge of color descriptor in MPEG-7, it designs image color descriptors that accord with MPEG-7 standard, and develops a demo version of CBIR system that supports MPEG-7 standard with the Visual C++ and the class library that supports MPEG-7.The experimental results show that, by adopting the MPEG-7 color descriptor, the retrieval speed can be greatly improved without damaging the accuracy. Because the system employs standard color descriptors, they can be reused by other applications that support MPEG-7. Meanwhile, the system can also use the descriptors generated by other applications that accord with MPEG-7, and so the system can be easily extended and the descriptors can also be more efficiently reused.
|
PDF Full Text Request