Infrared and Visible Image Fusion Using Improved QPSO-PCNN Algorithm

Abstract

This paper constructs a modified adaptive pulse coupled neural network (PCNN) model applied to infrared and visible image fusion. The PCNN model is improved by the quantum-behaved particle swarm optimization (QPSO) to set the parameters automatically. Two source images are processed by the QPSO-PCNN model respectively. Two hybrid fitness functions are proposed for QPSO algorithm. The first hybrid fitness function is made up of three evaluation criteria, spatial frequency (SF), average gradient (AG), and image entropy (EN). The other one uses mutual information based on image complex matrix (QCSVD_MI) instead of EN. Then, the fused image is generated by comparing the pulse output matrix of the two source images based on the judgment threshold. Finally, the proposed method is tested and verified with four pairs of infrared and visible images. Furthermore, the performances of different methods are judged with QCSVD_MI, EN, SF, AG and standard deviation (STD). Based on the experimental results, the proposed method is proved to be more suitable for complex infrared and visible image fusion. Especially, the method with fitness function based on QCSVD_MI, AG, and SF shows much better performance for relatively complex images. Because QCSVD_MI extracts more sensitive information of human visual system (HVS) which is more suitable for human eye observation.