Image Retrieval Algorithms For Content-based Web Image Search Engines

Based on the content of the image search engine (CBIR) is to search images which userwants according to the characteristics from images database. It has very important theoreticalsignificance for content-based search engine that studying on more effective image retrievaltechnology, since it could improve customer’s satisfaction on the image search engine. Thisarticle will use dynamic adjusting the weights area-weighted information entropy algorithm toextract image features. Based on research of relevant feedback, multi-feature fusion relevantfeedback algorithm has been achieved. On the basis of image feature extraction and imagesearch engine relevant feedback techniques, explore and research image feature extraction andrelevant feedback techniques suitable for content-based image search engine technology, anddevelop content-based search engine system. The major work includes.Existing image feature extraction algorithms have been analyzed, considering the fixedweights area-weighted information entropy extraction algorithm does not meet therequirements of complex digital image system, on the basis of area-weighted informationentropy, the dynamic adjusting the weights area-weighted information entropy has been putforward. Use the dynamic adjusting the weights area-weighted information entropy as theimage’s descriptor to indicate the image features, query the source image in the imagesegmentation based on analysis and research to determine the weights for each region of theimage based on image color space distribution, the dynamic adjusting the weightsarea-weighted information entropy can been achieved. Tests show that the precision rate toextract image features with the dynamic adjusting the weights area-weighted informationentropy algorithm is much6%than with the area-weighted information entropy algorithm.Relevant feedback techniques to the existing image search engines have been focus onthe multi-feature fusion relevant feedback algorithm. On the basis of the dynamic adjusting the weights area-weighted information entropy, with a combination of image HSV colorspace average color model of feature extraction algorithm, the to dynamically adjust theweights of the area-weighted information entropy algorithm proposed in this paper. Theaverage color model color feature extraction algorithm, the multi-feature fusion relevantfeedback techniques have been proposed based on the dynamic adjusting the weightsarea-weighted information entropy and the average color feature extraction algorithm. Takinginto account the image in complex digital image library does not use single image descriptor,according to the diversity of images, use the relevant feedback techniques. This method isapplied to content-based image search engine, experiments show that using the multi-featurefusion relevant feedback algorithm, after the user submit to retrieve images, the system isdesigned to extract the different characteristics of the image, image retrieval results based onimage feature vector can been achieved, the image results is constantly arranged according tothe relevant feedback technique which makes multi-feature extract relevant feedback retrievalaccuracy rate is much32.4%than the average color feature extraction algorithm,8.4%thanthe dynamic adjusting the weights area-weighted information entropy algorithm.Based on the above research. Our team design and implement a content-based Webimage search engine V3.0system. The system uses some technologies such as adjusting theweights area-weighted information entropy to extract image features, relevant feedbacktechniques, JNI and so on. The results show that the content-based Web image search engineV3.0system has well improved in performance and retrieval effects.