Content-Based Image Retrieval System Research
|School||Beijing University of Posts and Telecommunications|
|Course||Pattern Recognition and Intelligent Systems|
|Keywords||Content-based Image Retrieval Feature extraction Semantic links Relevance feedback|
Content-based image retrieval technology has become a large-scale multimedia database efficient management and retrieval tasks in an emerging technology in recent years in the field of multimedia information retrieval research focus . Its good prospects for the development and application of the increasing demand will continue to promote the development of this technology . The so-called content-based image retrieval (Content Based Image Retrieval, referred CBIR), also known as image content retrieval technology (Query By Image Content, referred QBIC), is the use of image content description information, such as color, shape, texture , semantic , etc. characteristics similar to the legend on the user's query image retrieval . This article discusses the content-based image retrieval technology research background and significance as well as its wide range of applications ; departure from the focus of the study , classification summarizes the research status CBIR technology , introduces some existing typical CBIR system prototyping : and comprehensive discussion of the content-based image retrieval key technologies , including: image content description method , feature similarity measure , features indexes, query mode system design , relevance feedback , CBIR retrieval criteria. In this technique, basic theoretical research , this paper focuses on proposed and implemented a CBIR system solutions , and to the system made ??a scientific evaluation of retrieval performance . Include: First , we use the SIFT descriptor intensive sampling method to extract the underlying characteristics of the image , through the \embedded image feature description, thereby obtaining a stable and more fulfilling human perceptual characteristics of image content characterization : Secondly , the paper through the use of class histogram intersection method to measure the similarity between image features ; Third , we use the label of positive and negative feedback graph group feedback mechanism , allowing the system from the user feedback explore their true intentions , in order to optimize retrieval effectiveness ; Finally , this paper two image database of the system was tested on the search results are scientific analysis and evaluation, and indicate the system needs further improvement, and future research directions.