Image Retrieval Optimize Algorithm Based On Feedback Verification

Advances in multimedia and internet technology have led to tremendous growthin very large image databases. The management of image database and how toretrieve image from those images becomes an especially challenge in image area.Text-based image retrieval work is by using the image captions, while themanual image annotation is time-consuming and therefore difficult to meet thedemand for huge image database. The semantic-based image retrieval which aims atdiscovering some high-level semantic meaning within an image. The cause of thesemantic gap is the failure of extracting real semantics from an image and the query.Thus the main obstacle to realize real semantic-based image retrieval is that semanticdescription of image is difficult. Content-based image retrieval (CBIR) is theapplication of computer vision techniques to the image retrieval problem. CBIRmeans that the search analyzes the visual contents that can be derived from the imageitself, which has become the main method for visual information retrieval.In order to make the retrieval images to be more in accordance with humanvisual, there are many works on CBIR. And the related studies can be divided intotwo groups: image visual feature extraction and the other is the relevant feedback forCBIR. The former category can fully express visual features in the image informationthrough extraction, selection and index. The latter through the users’ feedback toretrive images, and therefore improve performance of the retrival system.In this paper, we study the key issues of content-based image retrieval, and themain contents of research and innovation include:(1) The image feature extraction based on lifting scheme. Multiresolutionapproaches provide a powerful tool for image processing. And motivated by theadvantages of lifting scheme in the realization and implementation for wavelettransform, we extract the image feature with the lifted wavelet coefficient. In thispaper, we study the M-band lifting scheme for image feature extraction. Bycomparison the influences of different subbands on image visual features, we foundthat compared with the traditional2-band wavelet, the3-band lifting scheme will givemore high frequency subband information and can describe the texture informationmore accurately. And the weighted feature with the genetic algorithm is in theaccordance with the human visual perception. (2) Image retrieval optimize algorithm based on feedback verification. It is wellknown that the relevance feedback technology can improve the precision of the imageretrieval. However most of the feedback system depends on the user interaction,which make against the automatization. In this paper, we propose an efficientfeedback verification by adding a new retrieval program. In this system, the retrievalresults will be used as the query image and do retrieve again. And with this crossverification, we determine the retrival images. This process does not depend on userinvolvement or machine training.The experimental results show that the proposed feature extraction method cannot only greatly improve the image retrieval performance but also good for hardwareimplementation. Image retrieval optimize algorithm based on feedback verificationimprove the retrieval accuracy without user participation,which is of greatsignificance for the intelligent realization of image retrieval.