多分块主色与权值矩阵反馈的图像信息检索系统

图像检索是信息检索的重要内容。为了提高基于内容的图像检索效率．在主色调较明显的图像检索中，提出多分块策略算法；多分块及分块匹配便于控制检索的粒度．以及定位表现内容的主题画面。在此基础上加入分块的权值反馈进行多次检索可捕获用户意图．提高检索精度。在分析分块策略、颜色空间选取、矢量量化改进，权值矩阵更新等技术基础上，实现检索系统原型并进行对比实验。在“媒体眼中的广州”主题新闻数据库中的检索应用表明．多分块主色结合相关反馈相比全局检索和简单的分块直方图累加检索方法．提高图像检索的查准率．

Image Information Retrieval System Based on Multi-Block Primary Colors and Weight Matrix Feedback

Image retrieval is an important part of information retrieval, in order to improve the efficiency of content-based image retrieval, proposes using multi-block strategy algorithm in the image re- trieval with obvious primary color. Using multi-block and block matching is easy to control the retrieval granularity and to locate the topic screen. On this basis the block＇s weight feedback of retrieving for times can capture the user＇s intent and improve the retrieval accuracy. Based on analysis of multi-block strategy, color space selection, improves vector quantization and weight matrix updating, realizes a prototype of retrieval system and a comparison experiment. The data show that the primary colors based on multi-block with feedback retrieval compares to simple global histogram retrieval and cumulative histogram of sub-block retrieval method can improve the retrieval precision.