Recent progress in image denoising: A training strategy perspective

Abstract

Image denoising is one of the hottest topics in image restoration area, it has achieved great progress both in terms of quantity and quality in recent years, especially after the wide and intensive application of deep neural networks. In many deep learning based image denoising models, the performance can greatly benefit from the prepared clean/noisy image pairs used for model training, however, it also limits the application of these models in real denoising scenes. Therefore, more and more researchers tend to develop models that can be learned without image pairs, namely the denoising models that can be well generalised in real‐world denoising tasks. This motivates to make a survey on the recent development of image denoising methods. In this paper, the typical denoising methods from the perspective of model training are reviewed, the reviewed methods are categorised into four classes: the models need clean/noisy image pairs to train, the models trained on multiple noisy images, the models can be learned from a single noisy image, and the visual transformer based models. The denoising results of different denoisers were compared on some public datasets to discover the performance and advantages. The challenges and future directions in image denoising area are also discussed.

ConvNeXt-ST-AFF: A Novel Skin Disease Classification Model Based on Fusion of ConvNeXt and Swin Transformer

“…The presence of artifacts, such as body hair and blood vessels, in skin disease images may have a negative impact on the model's classification performance [30]. Image denoising is an effective method for reducing artifacts in images, [31]. When using traditional wavelet-based image denoising methods, it's necessary to first determine the type of noise present in an image, such as Gaussian noise or salt-and-pepper noise, and then choose specific wavelets for denoising based on the noise type.…”

Image restoration with group sparse representation and low‐rank group residual learning

“…First of all, it assumes that patches within the same group share the same truth representation, ignoring the specificity of patches. In other words, Equation (20) achieves the desirable performance if and only if patches within the same group are homogeneous. Actually, patches groups can be further divided into sub-groups, where each sub-group has a unique representation, indicating that there are multiple representations for the original patch group.…”

“…Actually, patches groups can be further divided into sub-groups, where each sub-group has a unique representation, indicating that there are multiple representations for the original patch group. Second, the relation among patches of the same groups is also neglected since Equation (20) only depicts the distance between patches and centers of groups decreasing the performance and interpretability of patterns. Furthermore, Equation (20) employs the weighted linear function to obtain B i , where weights of patches are difficult to select since the relative importance between them and exemplars is hard to measure.…”

“…where the W i ∈ ℝ m×m is a weight matrix. Equation ( 21) avoids sharing the unique center in the residual model in Equation (20), where multiple centers are allowed. In this case, the relations among patches within the same groups are explicitly exploited, which provides a better strategy to characterize and model the truth representation of patches.…”

“…Evidence demonstrates that image priors are the foundation for image restoration, including total variation (TV) [5][6][7], sparsity [2,8], low-rank [9][10][11], and deep image prior [12][13][14][15][16][17][18][19][20]. Particularly, sparsity prior is considered as one of the most remarkable for natural images [2,8,[21][22][23][24].…”