Abstract: An improved dual-channel pulse coupled neural network ( PCNN) is proposed for image fusionto address the challenges of manually setting numerous parameters and processing only one image at a timein traditional PCNN. The enhanced dual-channel model introduces a Gaussian filtering mechanism thatcombines surrounding neurons in a Gaussian distribution , eliminating the need for manually setting synapticweights and link amplification coefficients. Additionally, fractal dimension is used to estimate all other pa-rameters , fully achieving parameter adaptation. As a dual-channel model, it is capable of processing twoimages simultaneously. Two multi-source images are first decomposed into multi-scale representations usingthe non-subsampled contourlet transfom( NSCT ), yielding high-frequeney and low-frequeney sub-bandimages. The high-frequeney sub-bands are fused using the newly proposed improved dual-channel PCNNwhile the low-frequency sub-bands are fiused using an energy-based method. Finally, the fused low-fre.quency and high-frequeney sub-bands are inverse-transformed using NSCT to obtain the final fused image.Experimental results demonstrate that the proposed method offers significant advantages in terms of both vis.ual quality and objective evaluation metrics.

Key words: image fusion, non-subsampled contouret transform, improved dual-channel pulse coupled
neural network