Research And System Implementation On Content-Based Image Retrieval (CBIR)

With the development of database system and computer vision, image retrieval technique becomes a very active research field gradually. The content-based image retrieval (CBIR) is different from the traditional text-based image retrieval. It is a fuzzy inquiry technology in fact. Through extracting the characteristic of the image, we may find out the image that is most close to the enquiring requirement in the characteristic space. The issue involves two aspects: the extraction of characteristics, and the measurement of the distance between two characteristics vectors, i.e. the problem of characteristic match.In addition, it finally depends on the actual users to determine whether the two images are similar or not despise different adoption of characteristics and estimation of distance. Different users have different standards to the similar judgment of the image too. Hence, relevant feedback technology is advanced and applied to the image retrieval. The goal of it is to study from mutual course between the users and the inquiry system to find and catch their actual inquiry intentions, and revise the systematic inquiry tactics, thus, get the inquiry results as identical as possible to the users' actual demands. Because it could revise the systematic inquiry tactics real-timely, adaptive function has been added to the image retrieval system.The paper carries on the deep discussion and studies on the above three respects of the technology involved in CBIR system, and chosen suitable algorithm, designed and developed a CBIR experimental system. Chapter two of this paper is a key part of the full text, and discussed the extraction and expression of the color, texture, and shape characteristic separately. Main research contents and achievement have:First of all, a great deal of methods of extraction and expression about the color, texture and shape characteristic are summed up and summarized in detail. Secondly, we experiment the validity of some characteristics, for example, to the color characteristic, we compare with the validity of color histogram in different color space, and go on the contrast to Pieces of Multi-resolution main color index, and provide experiment demonstration finally.Based on the experiment, we collecte and combine the effective characteristic, and improve the algorithms in efficiency and accuracy. For example of shape characteristic, we chose eccentricity, compactness, shape invariant quadrature, andFourier descriptors to combine characteristic vector, and carry on the experiment and make better result.The third chapter mainly discusses the method of characteristic match, include: L1 and L2 distance, Mahalanobis distance, quadratic form distance and histogram intersection. In addition, characteristic match methods used in CBIR experimental system are presented.The innovation of the paper lies in chapter four. At first, we chose the simplest and valid method in numerous relevant feedback theories, and according to it, we design a simple and friendly inquiry interface. Chapter five mainly discussed the design and implementation of the experimental system, and provid the demonstration of systematic operation result. Chapter six is a summary of the full text, carry on objective appraisal to the author's work with conclusive language, and expect the development trends in this field.