中文摘要：

为分析比较基于内容的图像检索（content-based image retrieval,CBIR）系统中各环节算法在应用于刑侦现勘（criminal scenes investigation,CSI）图像数据库时的有效性,在CBIR系统中,使用颜色特征-HSV直方图、纹理特征-三层小波分解统计特征和颜色纹理的融合特征作为特征提取算法,以欧氏距离和街区距离作为相似性度量分别在CSI图库中进行仿真测试。在COREL图库中进行类似测试,并以查准率作为有效性的衡量标准进行对比。结果显示,在CSI库中,使用颜色特征有相对高的查准率,但CSI库的平均查准率均低于COREL库,此外使用街区距离可获得较高的查准率,这表明,因CSI数据场景复杂及目标多有损毁,现有图像特征提取算法有效性低,而作为相似性度量算法,街区距离相较于欧氏距离能更好地表示CSI图像间的相似性。

英文摘要：

In order to analyze the effectiveness of the existing content-based image retrieval（CBIR）algorithms when applied to crime scene investigation（CSI）image database,different image features are tested including color feature（the histogram of the HSV space）,texture feature（the statistic feature of three layer wavelet decomposition）,and the fusion of the color feature and texture feature.In addition,Euclidean distance and city block distance are used as the similarity measures.Using precision as the effectiveness evaluation,same methods are also tested on COREL database as comparison.Experimental results show that,in CSI database,color feature can provide higher precision than the other features.However,the average precision on CSI database is lower than that on COREL.Results also show that city block distance leads to higher retrieval precision.It can be concluded that,due to the complex background and often damaged targets in CSI images,the existing CBIR algorithms display low efficiency on CSI image database;and that for CSI image similarity measure,city block distance works better than Euclidean distance.