Multimodal biomedical image fusion method via rolling guidance filter and deep convolutional neural networks

Abstract Recently, numerous image fusion algorithms have been proposed and widely used in the biomedical field. However, most existing algorithms have low luminance, blurry edges and unclear details. To overcome these weaknesses, a multimodal biomedical image fusion method via rolling guidance filter and deep convolutional neural network is proposed in this paper. To enhance image edges and details, the VGG network is utilized. Our fusion algorithm includes three steps. First, the rolling guidance filter is utilized to extract the base and detail images. Second, the perceptual images are extracted by the VGG network to improve the fusion quality. Finally, the base, detail and perceptual images are fused by three different fusion strategies. Specially, the image decomposition parameters are selected based on experiments in order to extract appropriate texture and structure images. Additionally, the normalization operation is applied to the perceptual images that extracted by the VGG network for removing the feature differences and noise. Extensive experiments demonstrate that our proposed algorithm has better fusion results and objective metrics than the reference fusion methods.