Research And Implementation Of Content-based Image Retrieval System

With the rapid development of Multi-media technology,Internent of things,Cloud computing,Social networks and Big data, vast number of digital images are produced. How to find required images quickly from big datasets has become the hottest and most challenging problem.In order to solve such problem, Content-base Image Retrieval(CBIR) came into being.Image is a kind of description and record of objective world and the information described by image is far beyond language and words. According to statistics analysis, external information which human has receivedfrom visual information is more than 70%.CBIR is a retrieval system that can search relative images from database or network based on sample image of of user specified,features extracted from color, texture and shape. In CBIR, the query strategy can be adaptively adjusted.This paper introduces the research background, literature and developing trend of content-based image retrieval.Key technologies of content-based image retrieval is also be described in detail. Based on the research of relevance feedback technology, Content-based image retrieval system platform is designed and realized.The platform is divided into menu control area, samples area, sample selection area and results display area. After finished the platform, sample image adaptive selection, feature extraction, relevance feedback, and long-term learning are accomplished. Based on the image retrieval platform, two efficient image retrieval algorithms are studied and realized.Aiming at the pratical problem of large amount of calculation and easy to cause error matching of rectangle template, A fast image retrieval algorithm based on one loop and eight radiation direction. The algorithm is come of the topology of Beijing city traffic network which is composed of main road,supplemented by radioactive road and then can cover the entire Beijing area.In the algorithm,Featuresof two images are first extracted, and then similarity is computed by using Euclidean distance. Expermential results in 5000 Corel images show that the algorithm can greatly reduce retrieval speed without significantly reducing the accuracy.Considering retrieval accuracy is very low by using traditional image retrieval algorithm, relevance feedback technology and the thought of user retrieval concept are introduced into image retrieval. User retrieval concept is established through historical retrieval records and then we can establish one to many samples query algorithm. Experimental results show that the algorithm can greatly improve the accuracy of the image retrieval.