Enhanced Frequency Fusion Network with Dynamic Hash Attention for image denoising

A learnable full-frequency transformer dual generative adversarial network for underwater image enhancement

“…For this reason, influenced by octave convolution (Chen et al, 2019), Li et al (2021) use downsampling convolution to decompose mixed feature representations of images, and although they are able to decompose different frequency domain features, there is still unknown information loss due to the random loss of information features from convolution. In a similar operation, Jiang et al (2023) use dilated convolution to decompose a mixed feature representation of the image, which also suffer from random loss of information (which could not be quantified). The above studies show that the convolution approach is effective in decomposing image information, so what is the best way to do it without information loss?…”

“…DWT can represent the entire spatial frequency domain of an image and local spatial frequency domain features. In addition, scholars proposed dilated convolution filtering transform (DCFT) (Jiang et al, 2023) and up-/downsampling sampling filtering transform (USFT) (Li et al, 2021), both of which are difficult to measure quantitatively as the information through the convolution or up-/downsampling sampling will randomly lose high-frequency signals. Figures 4A-C show that different decomposition methods can separate low-, medium-, and high-frequency image differences to some extent, with the DWT being able to obtain better separation, while the convolution and up-/downsampling sampling filter transforms designed by the researcher are able to separate image differences in high-frequency images, and our proposed decomposition method can obtain better separation in each frequency domain.…”

The Medical Image Denoising Method Based on the CycleGAN and the Complex Shearlet Transform

“…For example, Z. Lyu et al constructed the "NSTBNet" model based on the non-subsampled shearlet transform and a broad convolutional neural network to remove spatially variant additive Gaussian noise [36], C. Gu et al combine the GAN and LSTM models for 3D reconstruction of Lung Tumors from CT Scans [37], Q. Song et al proposed the multimodal sparse transformer network (MMST) to remove the external noise in the task of the automatic speech recognition by using the mechanism of sparse self-attention [38] and B. Jiang et al constructed the so-called "EFFNet" model for image denoising by enhancing the transformed frequency features with dynamic hash attention [39]. Inspired by the above work, a novel image denoising method based on the complex shearlet transform and the cycle-consistent adversarial networks (CycleGAN) is developed to improve the denoising performance.…”