Research On Image Retrieval Algorithm Based On Multi-view Fusion

In the era of multimedia technologies, it has become more and more difficult for users to distinguish useful information from massive databases. The rapid growth of image number puts forward a challenge for people to retrieve image from vast amounts of information. The urgent needs require an efficient tool to help users meet their demands accurately and effectively as much as possible. Content-based Image Retrieval(CBIR) system can provide similar images according to their similarities when the system obtains an image. Generally speaking, the traditional method of image retrieval describes information by means of intuitive low-level features such as color, texture and shape. Research demonstrates that a single feature can't fully describe the image characteristics. So, in view of the image retrieval, this paper puts forward a multiple feature(i.e., multiple view) method for feature extraction and image retrieval, a method of high dimension characteristic of linear dimension reduction and a weighted layout descriptor based on region of interest. Research results are as follows:(1) In the process of image feature extraction, features are often high-dimensional. High-dimension features not only increase system's burden, affect the accuracy of characteristic, but also produce the serious redundancy. In this paper, we combine with the characteristics of coefficients of wavelet transform and give different weights to feature vectors on the basis of the principal component analysis method. So the feature vectors can present image information as much as possible.(2) With the analysis of the single characteristic of low-level features, we combine with characteristics in the various views. It takes advantage of useful characteristics at the same time, in order to solve the shortage problem of the single feature information.(3) Region based Image retrieval adds spatial layout information on the basis of content-based image retrieval technology. The paper proposes a weighted spacial method based on region of interest. System automatically defines the region of interest by analyzing color and texture features of image, and gives different weights to different locations of the regions. It highlights the voice of region of interest, conforms to the human visual characteristics, and provides the user better search experiments.