The Research Of Content-Based Image Retrieval System

With the development of multimedia, content-based image retrieval is becoming an important part in our everyday life and image retrieval. It could solve many problems existing in traditional text-based retrieval system, such as time-consuming, bad description and large workload. More and more focus is set on content-based image retrieval. Image data is very huge and it increases very drastically. How to retrieve information from image database is a crucial problem in the content-based image retrieval. Besides, color, texture, shape and spatial relation are the basic features of an image. Since it is very difficult to build concrete function relation between these basic features and semantic features, there is semantic gap between them and image retrieval is seldom satisfactory.The paper makes image retrieval on wavelet compression domain instead of original image data domain, and adopts the retrieval method based on feedback to identify similar images, thus offsets the semantic gap between basic features and semantic features; it puts those images wanted into image set and memorizes them, thus could reduce the range of retrieval, cut down computation, shorten retrieval time and improve retrieval precision; it combines memory-based learning with three features of an image to improve the recall and precision of the retrieval. The paper builds a three-layer image database, and realizes the transfer, computation and storage in simple format instead of complex format by the use of relation between image information and image features.