Detection of concealed weapons is an increasingly important problem for both military and police since global terrorism and crime have grown as threats over the years. This work presents two image fusion algorithms, one at pixel level and another at feature level, for efficient concealed weapon detection application. Both the algorithms presented in this work are based on the double-density dual-tree complex wavelet transform (DDDTCWT). In the pixel level fusion scheme, the fusion of low frequency band coefficients is determined by the local contrast, while the high frequency band fusion rule is developed with consideration of both texture feature of the human visual system (HVS) and local energy basis. In the feature level fusion algorithm, features are exacted using Gaussian Mixture model (GMM) based multiscale segmentation approach and the fusion rules are developed based on region activity measurement. Experiment results demonstrate the robustness and efficiency of the proposed algorithms.
展开▼
