School of Electronic and Information Engineering, Lanzhou Jiaotong University
Project supported by the National Natural Science Foundation of China (Grant Nos. 61861025，61562057，61761027，51669010) . Project supported by the Program for Changjiang Scholars and Innovative Research Team in University of Ministry of Education of China (Grant No. IRT_16R36). Project supported by Key Laboratory of Opto-technology and Intelligent Control, Ministry of Education(Lanzhou Jiaotong University) Open topic (Grant No. KFKT2018-9). Project supported by Lanzhou Talent Innovation and Entrepreneurship Project (Grant No. 2018-RC-117).
In order to solve the problem of unprominent features and low image contrast in infrared and visible image fusion, a hybrid model-driven fusion algorithm is proposed. First, the latent low-rank representation model is used to extract the base subband, significant subband and sparse noise subband of infrared and visible images respectively. Then, the base subbands are decomposed into low frequency coefficient and high frequency coefficient by using the non-subsampled shearlet transform modle, and the low frequency coefficient is accurately fitted by dictionary learning and sparse representation, while the high frequency coefficient is selected by local window combined with logic weighting. The significant subbands were fused by adaptive weighting of regional energy ratio threshold. Finally, the low frequency coefficient and high frequency coefficient after fusion were reconstructed in the first order to obtain the fusion base subband, while the sparse noise subband was discarded, and the fusion image was obtained by combining the fusion significant subband with the second-order reconstruction. The experimental results show that this algorithm can obtain the fusion image with rich information and relatively clear, which is feasible. The fusion result has high contrast and significant target contour, which can improve the recognition of the scene and has validity.