基于特征融合的刑侦图像数据库检索算法研究

项目名称： 基于特征融合的刑侦图像数据库检索算法研究

项目编号： No.61202183

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 刘颖

作者单位： 西安邮电大学

项目金额： 22万元

中文摘要： 刑侦图像和其它类型图像相比具有极大的特殊性，刑侦图像数据库的检索也具有非常强的针对性和目的性，如何构建适用于刑侦图像数据库检索的算法是我国公安系统急迫需要解决的问题。鉴于此，本项目根据刑侦图像的特性，研究适用于公安系统的刑侦图像数据库检索算法。主要研究内容包括：(1)分析实际刑侦图像的全局特征和局部特征，提出有效的刑侦图像视觉特征提取算法；(2)利用刑侦图像的语义层次结构，将刑侦图像语义模板和判决树技术相结合，有效地提取刑侦图像的语义特征；(3)通过实验数据，分析刑侦图像的视觉特征和语义特征在图像检索中的作用，提出适用于刑侦图像检索的特征融合算法，达到进一步提高检索性能的目的。 公共安全是我国社会安全稳定的基石。本项目的研究成果将会进一步提高我国警方的工作效率，节省大量的人力、物力和财力，因此本项目的研究不仅具有重要的理论价值，也具有极大的实用价值。

中文关键词： 现勘图像；轮胎花纹；特征提取；语义学习；检索效率

英文摘要： Criminal investigation image has its specific characteristics compared with images in other domains, and criminal investigation image database retrieval algorithms have to be designed based on these characteristics. Such algorithms are of urgent need in public security system in our homeland. This project aims to study on image retrieval algorithms based on the characteristics of criminal investigation images in public security system. The project includes three parts. The first is to find effective methods to extract visual features of criminal investigation images including both global features and regional features, based on an analysis in real-world image data. The next step is to obtain image semantics from visual image features by leveraging the semantic hierarchy of criminal image concepts and combining image semantic templates with decision tree learning. An understanding in the performance of both visual features and image semantics in criminal image database search can be obtained by analyzing the experimental results. Based on this, the third task in the project is to fuse visual image features and image semantics in order to further improve the performance of criminal image database search. Public security is the cornerstone of social stability in our country. The research in this project will help

英文关键词： Crime Scene investigation images；tire patterns；feature extraction；semantic extraction；retrieval performance