The Research And Implementation Of The Content-based Image Retrieval System

The popularity and application of Internet technology has led to significant development of social networking and big data technologies,and a large amount of digital image information has been produced.How to quickly and accurately match an image from the current massive multimedia library to find out the image resource information required by the user,the image retrieval has become a challenging research topic.Aiming at the traditional text annotation retrieval,this paper uses the visual characteristics of images to carry out content-based image retrieval research to realize "searching by graph".This paper takes the image of leaves and flowers as the research object.The key technologies of content-based image retrieval technology are studied,including color space,underlying feature description,image similarity measure,evaluation criteria of retrieval performance and feedback technology.The underlying feature description includes color,texture and shape features;the image similarity measure includes Euclidean distance,quadratic distance and intersection method;the evaluation criteria of retrieval performance include recall rate,precision rate and retrieval time;feedback technology includes feature weight and machine learning.In the in-depth study of the underlying feature description,color histograms,color moments and cumulative histograms are used on the color features,gray-scale co-occurrence matrices are used on the texture features,and a binarized segmentation is proposed on the shape features,the method of performing Hu invariant moment transformation.Aiming at the shortcomings of single feature retrieval,this paper proposes a comprehensive feature(color and shape)retrieval method,and adds relevant feedback technology to improve the retrieval performance of the system.Based on the previous algorithm research,an image retrieval system was designed and implemented.The retrieval results of various image features were verified by experiments.It was proved that the system was simple and effective,and realized a variety of feature retrieval and comprehensive feature search with related feedback.Figure21;Table5;Reference 50...