Abstract

The medical image fusion methods based on spatial domain were the hotspot of the earliest research. However, spatial domain technology produces spectral distortion and spatial distortion of fused images. The proposed siamese network is one of the three models for comparing patch similarity in the convolution neural network (CNN) model. Because its two weight branches are the same, the feature extraction or activity level measurement methods of the source image are the same. This has certain advantages over the models of pseudo-Siamese and 2-channel, and the ease of training of the siamese model is also the reason why it is favored in fusion applications. After obtaining the weight map, the Gaussian pyramid decomposition is used, and the pyramid transform is used for multiscale decomposition, so that the fusion process is more in line with human visual perception. In addition, the localized similarity-based fusion strategy is used to adaptively adjust the decomposed coefficients. The algorithm combines the common pyramid based and similarity-based fusion algorithm with the CNN model to produce a superior fusion method.

Description