Image fusion method based on structure-based saliency map and FDST-PCNN framework

Image fusion has become an active and promising research topic in image processing. It provides an effective way to combine several source images to form a composite image with more detailed information than any one of the source images. An FDST-PCNN framework, which integrates finite discrete shearlet transform (FDST) with pulse-coupled neural network (PCNN), is proposed to possess a higher ability enhance fusion effects. We first propose a structure-based saliency (SBS) map to enhance the clear and important features in one image. The SBS map combines the depth information with the saliency information and could be a good representation of the most essential information of the source images. After multi-scale decomposition by the FDST, the SBS map of the source images and the modified-spatial-frequency of the subbands are both utilized to tune the PCNN neuron response and determine the fused coefficients in each subband. The experimental results on multi-focus and multi-sensor images verify the effectiveness of our proposed fusion method. Compared with other PCNN-based fusion methods, the proposed method achieves significant improvement in preserving detailed edge information and improving overall visual performance.