HFDDLap: High-Low Frequency Differentiation Dynamic Laplacian Pyramid Network for Image Super-Resolution

Investigating single image super-resolution algorithm with deep learning using convolutional neural network for chest digital tomosynthesis

The success of modern single image super-resolution (SISR) algorithms is inspired by the development of deep convolutional neural networks (CNNs). However, these CNN-based methods require considerable computation and complexity, making it impossible for these methods to perform real-time calculations in edge devices. Thus, lightweight model design has become a development trend in the super-resolution field, including pruning, quantization, and other methods. The 1-bit quantization is an extreme lightweight method which can reduce the calculation amount of the model in an extreme manner and is friendly to hardware such as edge devices. Most existing binary quantization approaches lead to a large information loss during forward propagation, especially in detailed color information (e.g., edge, texture, and contrast). The loss of color information makes modern binary methods unsuitable for SISR tasks. We think the loss occurs because these methods typically utilize a uniform threshold to quantize the weights and activations. Thus, in this article, we thoroughly analyze the difference between normal classification tasks and SISR tasks, and present a binarization scheme based on local means. The proposed method can maintain more detailed information in feature maps using dynamic thresholds during quantization. Specifically, each value in the full precision activations has a corresponding threshold during the quantization process, and those thresholds are determined by the full precision values of the surroundings. In addition, a gradient approximator is introduced to adaptively optimize the gradient for updating binary weights. We then verify the effectiveness of our method for training binary networks on several SISR benchmarks including VDSR and SRResNet. Experimental results show that the proposed method can outperform the state-of-the-art algorithms to obtain binary networks for image super-resolution with better peak signal-to-noise ratio (PSNR) values and visual quality.

Convolutional neural networks (CNN) have given rise to a new generation of video super-resolution (SR) technique. However, most existing CNN-based video SR algorithms treat the consecutive frames as a series of feature maps, just as the procedure performed in single image SR algorithms. We propose an end-to-end three-dimensional (3-D) CNN video SR framework. The input frames are considered as a cube in our framework. 3-D convolution is performed on it to extract features along spatial and temporal dimension. Image prior knowledge, such as optical flows, is introduced in reconstruction. A combination of mean square error loss and multiscale structure similarity index (MS-SSIM) loss is used to optimize the model. Experimental results show that the proposed method reconstructs high-resolution frames with more accurate and visually pleasant structures compared with state-of-the-art video SR algorithms. We also achieve comparable PSNR/SSIM results with less computation time.

The recent advances in deep convolutional neural networks (DCNNs) have convincingly demonstrated high-capability reconstruction for single image super-resolution (SR). However, it is a big challenge for most DCNNs-based SR models when the scaling factor increases. In this paper, we propose a novel Progressive Residual Network (PRNet) to integrate hierarchical and scale features for single image SR, which works well for both small and large scaling factors. Specifically, we introduce a Progressive Residual Module (PRM) to extract local multi-scale features through dense connected up-sampling convolution layers. Meanwhile, by embedding residual learning into each module, the relative information between high-resolution and low-resolution multi-scale features is fully exploited to boost reconstruction performance. Finally, the scale-specific features are fused to the reconstruction module for restoring the high-quality image. Extensive quantitative and qualitative evaluations on benchmark datasets illustrate that our PRNet achieves superior performance and in particular obtains new state-of-the-art results for large scaling factors such as 4 × and 8 ×.

Ultra-high-definition display technology is widely used in broadcasting, but there is a huge contradiction between its ultra-high-resolution content and short storage. Super-Resolution (SR) can effectively alleviate this contradiction. Recently, State-of-the-art image SR approaches leveraging Deep Convolutional Neural Networks (DCNNs) have demonstrated high-quality reconstruction performance. However, most of them suffer from large model parameters, which restricts their practical application. Besides, image SR for large scaling factors (e.g., ×8) is a tricky issue when the parameters diminish. To remedy these issues, we propose the Lightweight Multi-scale Aggregation Network (LMAN) for the image SR, which works well for both small and large scaling factors with limited parameters. Specifically, we propose a Group-wise Multi-scale Block (GMB) in which a group convolution is exploited for extracting and fusing multi-scale features before a channel attention layer to obtain discriminative features. Additionally, we present a novel Hierarchical Spatial Attention (HSA) mechanism to jointly and adaptively fuse local and global hierarchical features for high-resolution image reconstruction. Extensive experiments illustrate that our LMAN achieves superior performance against state-of-the-art methods with similar parameters and in particular for large scaling factors such as 4× and 8×.

Most of the state of the art super-resolution methods use deep networks with large filter sizes. Therefore, they need to train and store a correspondingly large number of parameters, thereby making their use difficult for mobile devices applications such as recognition of individuals from selfie images. To achieve an efficient super-resolution method, we propose an Efficient Single Image Super-Resolution (ESISR) algorithm, which takes into account a trade-off among the efficiency of the deep neural network, the size of the filters, and the sharpness of the images. To that end, the method implements a novel loss function based on the Sharpness metric. This metric turns out to be more suitable for recovering the quality of the eye images. Our method drastically reduces the number of parameters when compared with Deep CNNs with Skip Connection and Network (DCSCN): from 1,754,942 to 27,209 parameters when the image size is increased by a factor of 2 (x2), from 2,170,142 to 28,654 parameters when increased by 3 (x3), and from 2,087,102 to 64,201 parameters when increased by 4 (x4). Furthermore, the proposed method maintains the sharpness quality of the images.

Special Section on Enabling Technologies for Network-Based Applications

Image super-resolution based on half quadratic splitting

Deployment of deep convolutional neural networks (CNNs) in single image super-resolution (SISR) for edge computing devices is mainly hampered by the huge computational cost. In this work, we propose a lightweight image super-resolution (SR) network based on a reparameterizable multibranch bottleneck module (RMBM). In the training phase, RMBM efficiently extracts high-frequency information by utilizing multibranch structures, including bottleneck residual block (BRB), inverted bottleneck residual block (IBRB), and expand-squeeze convolution block (ESB). In the inference phase, the multibranch structures can be combined into a single 3 × 3 convolution to reduce the number of parameters without incurring any additional computational cost. Furthermore, a novel peak-structure-edge (PSE) loss is proposed to resolve the problem of oversmoothed reconstructed images while significantly improving image structure similarity. Finally, we optimize and deploy the algorithm on the edge devices equipped with the rockchip neural processor unit (RKNPU) to achieve real-time SR reconstruction. Extensive experiments on natural image datasets and remote sensing image datasets show that our network outperforms advanced lightweight SR networks regarding objective evaluation metrics and subjective vision quality. The reconstruction results demonstrate that the proposed network can achieve higher SR performance with a 98.1 K model size, which can be effectively deployed to edge computing devices.

The image super-resolution algorithm can overcome the imaging system's hardware limitation and obtain higher resolution and clearer images. Existing super-resolution methods based on convolutional neural networks(CNN) can learn the mapping relationship between high-resolution(HR) and low-resolution(LR) images. However, when the reconstruction target is a face image, the reconstruction results often have problems that the face area is too smooth and lacks details. We propose a guided cascaded face super-resolution network, called guided cascaded super-resolution network (GCFSRnet). GCFSRnet takes the LR image and a high-quality guided image as inputs, and it consists of a pose deformation module and a super-resolution network. Firstly, the pose deformation module converts the guide image's posture into the same as the low-resolution face image based on 3D fitting and 3D morphable model (3DMM). Then, the LR image and the deformed guide image are used as input of the super-resolution network. The super-resolution networks are formed by a cascade of two layers of networks, which extract different features. During the reconstruction process, the guide image can provide real facial details and help generate subtle facial textures. The cascade structure of a super-resolution network can gradually extract features and restore different levels of image details. The experimental results on the CASIA Web Face and CelebA datasets show that the proposed method can generate facial images with clear outlines and rich details, which are superior to other state-of-the-art methods such as SRResNet, SRGAN, VDSR, DBPN, etc.

Despite breakthrough advances in image super-resolution (SR) with convolutional neural networks (CNNs), SR has yet to enjoy ubiquitous applications due to the high computational complexity of SR networks. Quantization is one of the promising approaches to solve this problem. However, existing methods fail to quantize SR models with a bit-width lower than 8 bits, suffering from severe accuracy loss due to fixed bit-width quantization applied everywhere. In this work, to achieve high average bit-reduction with less accuracy loss, we propose a novel Content-Aware Dynamic Quantization (CADyQ) method for SR networks that allocates optimal bits to local regions and layers adaptively based on the local contents of an input image. To this end, a trainable bit selector module is introduced to determine the proper bit-width and quantization level for each layer and a given local image patch. This module is governed by the quantization sensitivity that is estimated by using both the average magnitude of image gradient of the patch and the standard deviation of the input feature of the layer. The proposed quantization pipeline has been tested on various SR networks and evaluated on several standard benchmarks extensively. Significant reduction in computational complexity and the elevated restoration accuracy clearly demonstrate the effectiveness of the proposed CADyQ framework for SR. Codes are available at https://github.com/Cheeun/CADyQ.

Adversarially trained deep neural networks have significantly improved performance of single image super resolution, by hallucinating photorealistic local textures, thereby greatly reducing the perception difference between a real high resolution image and its super resolved (SR) counterpart. However, application to medical imaging requires preservation of diagnostically relevant features while refraining from introducing any diagnostically confusing artifacts. We propose using a deep convolutional super resolution network (SRNet) trained for (i) minimising reconstruction loss between the real and SR images, and (ii) maximally confusing learned relativistic visual Turing test (rVTT) networks to discriminate between (a) pair of real and SR images (T1) and (b) pair of patches in real and SR selected from region of interest (T2). The adversarial loss of T1 and T2 while backpropagated through SRNet helps it learn to reconstruct pathorealism in the regions of interest such as white blood cells (WBC) in peripheral blood smears or epithelial cells in histopathology of cancerous biopsy tissues, which are experimentally demonstrated here. Experiments performed for measuring signal distortion loss using peak signal to noise ratio (pSNR) and structural similarity (SSIM) with variation of SR scale factors, impact of rVTT adversarial losses, and impact on reporting using SR on a commercially available artificial intelligence (AI) digital pathology system substantiate our claims.

In contrast to sophisticated means of visual su-per resolution (SR), not much work has been done in the tactile SR field. Existing tactile SR algorithms for taxel-based sensors mainly focus on enhancing the localization accuracy, and generally associate with a specific type of hardware, sometimes not applicable to generic taxel-based tactile sensors. Inspired by image SR, we investigate the tactile pattern SR in this paper, and present how to transform successful image SR schemes, e.g. Convolutional Neural Network (CNN) and Generative Adversarial Network (GAN) to serve the tactile SR. We propose two tactile SR models, i.e. TactileSRCNN and TactileSRGAN, and establish a new tactile pattern SR dataset for model learning. The ground truth of high resolution (HR) tactile patterns in the dataset is obtained via multi-sampling (i.e. overlapping reception) and registration of low resolution (LR) sensor. One key contribution of this research lies in achieving ×100 (from 3×4×4 to 40×40) times tactile pattern SR with a one-time tapping of 3-axis taxel-based sensor. Different from existing tactile SR algorithms which improves the localization accuracy of a single contact point, the proposed scheme can provide multi-point contact detection to robotic applications.

Recent studies have shown that the use of deep convolutional neural networks (CNNs) can improve the performance of single image super-resolution reconstruction (SISR) methods. However, the existing CNN-based SISR model ignores the multi-scale features and shallow and deep features of the image, resulting in relatively low image reconstruction performance. To address these issues, this paper proposes a new multi-scale fractal residual network (MSFRN) for image super-resolution. On the basis of residual learning, a multi-scale fractal residual block (MSFRB) is designed. This block uses convolution kernels of different sizes to extract image multi-scale features and uses multiple paths to extract and fuse image features of different depths. Then, the shallow features extracted at the shallow feature extraction stage and the local features output by all MSFRBs are used to perform global hierarchical feature fusion. Finally, through sub-pixel convolution, the fused global features are used to reconstruct high-resolution images from low-resolution images. The experimental results on the five standard benchmark datasets show that MSFRN improved subjective visual effects and objective image quality evaluation indicators, and is superior to other state-of-the-art SISR methods.

Deep convolutional neural networks (CNNs), trained on corresponding pairs of high- and low-resolution images, achieve state-of-the-art performance in single-image super- resolution and surpass previous signal-processing based approaches. However, their performance is limited when applied to real photographs. The reason lies in their train- ing data: low-resolution (LR) images are obtained by bicu- bic interpolation of the corresponding high-resolution (HR) images. The applied convolution kernel significantly differs from real-world camera-blur. Consequently, while current CNNs well super-resolve bicubic-downsampled LR images, they often fail on camera-captured LR images. To improve generalization and robustness of deep super- resolution CNNs on real photographs, we present a ker- nel modeling super-resolution network (KMSR) that incor- porates blur-kernel modeling in the training. Our pro- posed KMSR consists of two stages: we first build a pool of realistic blur-kernels with a generative adversarial net- work (GAN) and then we train a super-resolution network with HR and corresponding LR images constructed with the generated kernels. Our extensive experimental vali- dations demonstrate the effectiveness of our single-image super-resolution approach on photographs with unknown blur-kernels.