Aimed to better perform image reconstruction and enhance the processing of multi-scale feature representation and fusion of images,a new multi-scale feature fusion image reconstruction network model was proposed in this study.The model included iterative downsampling and iterative upsampling processes.The iterative downsampling process was completed by combining the Gaussian convolution kernel with the iterative operation rules of subsampling and Gaussian smooth filtering in the Laplacian pyramid residual control network model.The iterative upsampling process was achieved by using Laplace convolution kernels and second-order differential operation rules.The GJ-UNet deep learning network model achieved fine classification of image multi-scale semantic features through its encoder downsampling module,and applied deconvolution and convolution operation rules in the decoder upsampling module to standardize the processing of image multi-scale semantic features.The results showed that the proposed method can achieve high-precision feature extraction,and has stronger correlation with image feature fusion.The extracted image edge information is clearer and has lower relative noise information.The visual effect of the reconstructed image is basically the same as the original input image.This study is expected to be widely used in the field of computer image vision.
Compression-aware image reconstructionMultiscale feature of imageLaplace pyramid modelDifferential operationGJ-UNet deep learning network modelDice loss function
NETL
NSTL
万方数据
