Polarization Image Fusion Algorithm Using NSCT and CNN

Fusion is beneficial to improve the accuracy of target detection in polarization images. In this paper, we propose an image fusion algorithm that combines nonsubsampled contourlet transform (NSCT) and Convolutional Neural Network (CNN). First, the fast guide filter and pulse-coupled neural network (PCNN) are employed to eliminate the noise of the polarization angle images. Second, the degree of linear polarization and polarization angle images are fused to obtain the polarization characteristic image. Finally, the polarization characteristic image is fused with the intensity image by NSCT. The CNN network is involved to extract mutilayer features from the high-frequency components of NSCT. The zero-phase component analysis (ZCA) is applied to normalize these features. The fusion coefficients of the high-frequency subband fusion coefficients are obtained by calculating the activity energy map of feature image. The low-frequency subband fusion coefficients are obtained by the strategy of regional energy. The result image is reconstructed by inverse NSCT. The experimental results show that the average gradient of the fused image is 51.3% higher than that of polarized intensity image, while the spatial frequency is improved by 35.1%. This algorithm is superior to others in terms of subjective visual effect and objective evaluation parameters, which is suitable for target detection.