Concealed object recognition based on geometric feature descriptors

Millimeter wave (MMW) imaging is finding rapid adoption in security applications such as concealed object detection under clothing. A passive MMW imaging system can operate as a stand-off type sensor that scans people in both indoors and outdoors. However, the imaging system often suffers from the diffraction limit and the low signal level. This paper discusses real-time concealed object recognition based on geometric descriptors. The concealed object region is extracted by the multi-level segmentation method. A novel approach is proposed to measure similarity between a true object model and segmented binary objects. Principal component analysis (PCA) regularizes the shape in terms of translation and rotation. Size normalization provides scale-invariant property. A geometric feature vector is composed of several shape descriptors. The feature vector is invariant to scale, rotation, and translation, and tolerant to distortion. Classification is performed by means of measuring Euclidean distance between the mean feature vector of training models and the feature vector of the segmented object. Experiments confirm that the proposed method provides fast and reliable recognition of the concealed object carried by a moving human subject.