针对刑侦图像数量大、质量差、管理难的特点,采用了一种基于模糊分类理论对刑侦视频图像的场景进行分类的方法。首先对监控视频图像的场景进行人工多标记分类,然后对刑侦视频图像提取两种纹理特征(局部二值模式和小波纹理)并进行融合,最后采用模糊K-最近邻(K-nearest neighbor,KNN)分类器实现刑侦图像四种场景(车辆、行人、建筑和街道)的分类并得到监控视频数据库中图像的模糊不确定性。实验结果表明,隶属度充分反映了刑侦图像的内容,同时分类的正确率高达85%,初步达到了对刑侦视频图像自动分类管理的目的。
Based on the characteristics of criminal investigation images with large amount, poor quality and difficulty in man- agement, this paper studied the classification of criminal investigation image scenes by fuzzy classification theory. The first step was muhi-labeling the criminal investigation video images manually; the second step was to extract the texture features--local hinary pattern (LBP) and wavelet and fused them together; the last work was using the fuzzy KNN classifier to classify the criminal investigation images scene ( cars, people, buildings and streets) and got the fuzzy uncertainty of the images. The re- sults show that the memberships can represent the contents of the criminal investigation images effectively and the classification rate is up to 85%, preliminarily reaching the aim of automatic management of the criminal investigation images.