Infrared and visible light image fusion algorithm based on ResNet50 and non-separable wavelets
HAO Yuquan
Abstract：To address the limitation that inseparable wavelets decomposition fails to maximise high-frequency feature extraction in images,an infrared-visible light image fusion algorithm based on ResNet50 and inseparable wavelets is proposed.Firstly,infrared and visible light images are decomposed into low-frequency and high-frequency components using four-channel inseparable wavelets.Secondly,low-frequency subband images are fused via Delaunay interpolation and the maximum symmetric surrounding saliency method.Subsequently,the ResNet50 network extracts features from the high-frequency detail layer.The resulting feature map undergoes L1 regularisation,with a weight map generated via a maximum selection strategy to reconstruct a new high-frequency detail layer.Finally,the new reconstructed low-frequency and high-frequency images are combined to produce the fused image.By decomposing and reconstructing both frequency domains,this algorithm effectively preserves the source image's structural and detail information.The achieved fused image demonstrates satisfactory performance across both subjective and objective evaluation metrics.
Keywords：non-separable waveletsResNet50 neural networkimage fusionfusion rules
Publication Date：2025-10-25
Online Publishing Date：2025-11-24(First online date of this platform, not the publication date of the document)
Pages：8( 9-15,38 )