CBS安检图像中人体携带物特征提取与形状识别

Feature Extraction and Recognition of the Concealed Object Based on CBS's Human Body Image

基于全局分割与局部分割相结合的两层次阈值分割方法提取出人体携带物,在Hu不变矩基础上引入其他基于几何特征的分量建立了人体携带物特征表达向量模型,通过设计的三层BP神经网络模型实现对人体携带物的形状识别。实验表明:构建的形状特征表达向量和神经网络模型能有效应用于CBS图像中人体携带物的自动检测与识别;对手枪、管制刀具、扳子、钳子等携带物进行形状识别,准确率为90%。

Based on our double- level segmentation method including global segmentation and local segmentation for extracting concealed object,this paper proposes a new feature vector model composed of Hu invariant moments and other geometric- based feature vector components,and then adopts three- layer BP network model to identify concealed object under clothing. Experiments confirm that our method is effective for automatic detection and recognition of the concealed object under clothing,and the recognition rate of handguns,controlled knives,spanners and pliers is 90%.