Super-resolution Framework Research Articles

Towards Efficient Video Detection Object Super-Resolution with Deep Fusion Network for Public Safety

Video surveillance plays an increasingly important role in public security and is a technical foundation for constructing safe and smart cities. The traditional video surveillance systems can only provide real-time monitoring or manually analyze cases by reviewing the surveillance video. So, it is difficult to use the data sampled from the surveillance video effectively. In this paper, we proposed an efficient video detection object super-resolution with a deep fusion network for public security. Firstly, we designed a super-resolution framework for video detection objects. By fusing object detection algorithms, video keyframe selection algorithms, and super-resolution reconstruction algorithms, we proposed a deep learning-based intelligent video detection object super-resolution (SR) method. Secondly, we designed a regression-based object detection algorithm and a key video frame selection algorithm. The object detection algorithm is used to assist police and security personnel to track suspicious objects in real time. The keyframe selection algorithm can select key information from a large amount of redundant information, which helps to improve the efficiency of video content analysis and reduce labor costs. Finally, we designed an asymmetric depth recursive back-projection network for super-resolution reconstruction. By combining the advantages of the pixel-based super-resolution algorithm and the feature space-based super-resolution algorithm, we improved the resolution and the visual perception clarity of the key objects. Extensive experimental evaluations show the efficiency and effectiveness of our method.

Single depth map super-resolution via joint non-local self-similarity modeling and local multi-directional gradient-guided regularization

The development of consumer depth cameras makes it possible to acquire depth information of a scene in real-time. However, low resolution and low quality of a depth map has greatly constrained its applications. In this paper, we propose a novel framework for single depth map super-resolution, which considers local and non-local information jointly in the depth map. For the non-local constraint, group-based sparse representation is used to explore non-local self-similarity in the depth map. For the local constraint, a multi-directional gradient-guided regularization is proposed to describe the gradient of the depth map with spatially varying orientations. The former constraint contains the visual artifacts effectively, while the latter restores sharp edge and fine structure. Finally, the two complementary regularizers are jointly casted into a unified optimization framework, where a split Bregman-based technique is developed to tackle the optimization problem. Both quantitative and qualitative evaluations indicate that the proposed method can obtain better reconstruction performance compared with state-of-the-art methods.

Small Object Detection in Remote Sensing Images with Residual Feature Aggregation-Based Super-Resolution and Object Detector Network

This paper deals with detecting small objects in remote sensing images from satellites or any aerial vehicle by utilizing the concept of image super-resolution for image resolution enhancement using a deep-learning-based detection method. This paper provides a rationale for image super-resolution for small objects by improving the current super-resolution (SR) framework by incorporating a cyclic generative adversarial network (GAN) and residual feature aggregation (RFA) to improve detection performance. The novelty of the method is threefold: first, a framework is proposed, independent of the final object detector used in research, i.e., YOLOv3 could be replaced with Faster R-CNN or any object detector to perform object detection; second, a residual feature aggregation network was used in the generator, which significantly improved the detection performance as the RFA network detected complex features; and third, the whole network was transformed into a cyclic GAN. The image super-resolution cyclic GAN with RFA and YOLO as the detection network is termed as SRCGAN-RFA-YOLO, which is compared with the detection accuracies of other methods. Rigorous experiments on both satellite images and aerial images (ISPRS Potsdam, VAID, and Draper Satellite Image Chronology datasets) were performed, and the results showed that the detection performance increased by using super-resolution methods for spatial resolution enhancement; for an IoU of 0.10, AP of 0.7867 was achieved for a scale factor of 16.

Spatial and temporal super-resolution for fluorescence microscopy by a recurrent neural network.

A novel spatial and temporal super-resolution (SR) framework based on a recurrent neural network (RNN) is demonstrated. In this work, we learn the complex yet useful features from the temporal data by taking advantage of structural characteristics of RNN and a skip connection. The usage of supervision mechanism is not only making full use of the intermediate output of each recurrent layer to recover the final output, but also alleviating vanishing/exploding gradients during the back-propagation. The proposed scheme achieves excellent reconstruction results, improving both the spatial and temporal resolution of fluorescence images including the simulated and real tubulin datasets. Besides, robustness against various critical metrics, such as the full-width at half-maximum (FWHM) and molecular density, can also be incorporated. In the validation, the performance can be increased by more than 20% for intensity profile, and 8% for FWHM, and the running time can be saved at least 40% compared with the classic Deep-STORM method, a high-performance net which is popularly used for comparison.

Brain graph super-resolution using adversarial graph neural network with application to functional brain connectivity.

Brain image analysis has advanced substantially in recent years with the proliferation of neuroimaging datasets acquired at different resolutions. While research on brain image super-resolution has undergone a rapid development in the recent years, brain graph super-resolution is still poorly investigated because of the complex nature of non-Euclidean graph data. In this paper, we propose the first-ever deep graph super-resolution (GSR) framework that attempts to automatically generate high-resolution (HR) brain graphs with N' nodes (i.e., anatomical regions of interest (ROIs)) from low-resolution (LR) graphs with N nodes where N<N'. First, we formalize our GSR problem as a node feature embedding learning task. Once the HR nodes' embeddings are learned, the pairwise connectivity strength between brain ROIs can be derived through an aggregation rule based on a novel Graph U-Net architecture. While typically the Graph U-Net is a node-focused architecture where graph embedding depends mainly on node attributes, we propose a graph-focused architecture where the node feature embedding is based on the graph topology. Second, inspired by graph spectral theory, we break the symmetry of the U-Net architecture by super-resolving the low-resolution brain graph structure and node content with a GSR layer and two graph convolutional network layers to further learn the node embeddings in the HR graph. Third, to handle the domain shift between the ground-truth and the predicted HR brain graphs, we incorporate adversarial regularization to align their respective distributions. Our proposed AGSR-Net framework outperformed its variants for predicting high-resolution functional brain graphs from low-resolution ones. Our AGSR-Net code is available on GitHub at https://github.com/basiralab/AGSR-Net.

DL-MRI: A Unified Framework of Deep Learning-Based MRI Super Resolution.

Magnetic resonance imaging (MRI) is widely used in the detection and diagnosis of diseases. High-resolution MR images will help doctors to locate lesions and diagnose diseases. However, the acquisition of high-resolution MR images requires high magnetic field intensity and long scanning time, which will bring discomfort to patients and easily introduce motion artifacts, resulting in image quality degradation. Therefore, the resolution of hardware imaging has reached its limit. Based on this situation, a unified framework based on deep learning super resolution is proposed to transfer state-of-the-art deep learning methods of natural images to MRI super resolution. Compared with the traditional image super-resolution method, the deep learning super-resolution method has stronger feature extraction and characterization ability, can learn prior knowledge from a large number of sample data, and has a more stable and excellent image reconstruction effect. We propose a unified framework of deep learning -based MRI super resolution, which has five current deep learning methods with the best super-resolution effect. In addition, a high-low resolution MR image dataset with the scales of ×2, ×3, and ×4 was constructed, covering 4 parts of the skull, knee, breast, and head and neck. Experimental results show that the proposed unified framework of deep learning super resolution has a better reconstruction effect on the data than traditional methods and provides a standard dataset and experimental benchmark for the application of deep learning super resolution in MR images.

A Deep Multi-Frame Super-Resolution Network for Dynamic Scenes

The multi-frame super-resolution techniques have been prosperous over the past two decades. However, little attention has been paid to the combination of deep learning and multi-frame super-resolution. One reason is that most deep learning-based super-resolution methods cannot handle variant numbers of input frames. Another reason is that it is hard to capture accurate temporal and spatial information because of the misalignment of input images. To solve these problems, we propose an optical-flow-based multi-frame super-resolution framework, which is capable of dealing with various numbers of input frames. This framework enables to make full use of the input frames, allowing it to obtain better performance. In addition, we use a spatial subpixel alignment module for more accurate subpixel-wise spatial alignment and introduce a dual weighting module to generate weights for temporal fusion. Both two modules lead to more effective and accurate temporal fusion. We compare our method with other state-of-the-art methods and conduct ablation studies on our method. The results of qualitative and quantitative analyses show that our method achieves state-of-the-art performances, demonstrating the advantage of the designed framework and the necessity of proposed modules.

Deep Unsupervised Fusion Learning for Hyperspectral Image Super Resolution.

Hyperspectral image (HSI) super-resolution (SR) is a challenging task due to its ill-posed nature, and has attracted extensive attention by the research community. Previous methods concentrated on leveraging various hand-crafted image priors of a latent high-resolution hyperspectral (HR-HS) image to regularize the degradation model of the observed low-resolution hyperspectral (LR-HS) and HR-RGB images. Different optimization strategies for searching a plausible solution, which usually leads to a limited reconstruction performance, were also exploited. Recently, deep-learning-based methods evolved for automatically learning the abundant image priors in a latent HR-HS image. These methods have made great progress for HS image super resolution. Current deep-learning methods have faced difficulties in designing more complicated and deeper neural network architectures for boosting the performance. They also require large-scale training triplets, such as the LR-HS, HR-RGB, and their corresponding HR-HS images for neural network training. These training triplets significantly limit their applicability to real scenarios. In this work, a deep unsupervised fusion-learning framework for generating a latent HR-HS image using only the observed LR-HS and HR-RGB images without previous preparation of any other training triplets is proposed. Based on the fact that a convolutional neural network architecture is capable of capturing a large number of low-level statistics (priors) of images, the automatic learning of underlying priors of spatial structures and spectral attributes in a latent HR-HS image using only its corresponding degraded observations is promoted. Specifically, the parameter space of a generative neural network used for learning the required HR-HS image to minimize the reconstruction errors of the observations using mathematical relations between data is investigated. Moreover, special convolutional layers for approximating the degradation operations between observations and the latent HR-HS image are specifically to construct an end-to-end unsupervised learning framework for HS image super-resolution. Experiments on two benchmark HS datasets, including the CAVE and Harvard, demonstrate that the proposed method can is capable of producing very promising results, even under a large upscaling factor. Furthermore, it can outperform other unsupervised state-of-the-art methods by a large margin, and manifests its superiority and efficiency.

A New Super Resolution Framework Based on Multi-Task Learning for Remote Sensing Images.

Super-resolution (SR) algorithms based on deep learning have dominated in various tasks, including medical imaging, street view surveillance and face recognition. In the remote sensing field, most of the current SR methods utilize the low-resolution (LR) images that directly bicubic downsampled the high-resolution (HR) images as not only train set but also test set, thus achieving high PSNR/SSIM scores but showing performance drop in application because the degradation model in remote sensing images is subjected to Gaussian blur with unknown parameters. Inspired by multi-task learning strategy, we propose a multiple-blur-kernel super-resolution framework (MSF), in which a multiple-blur-kernel learning module (MLM) optimizes the parameters of the network transferable and sensitive for SR procedures with different blur kernels. Besides, to simultaneously exploit the prior of the large-scale remote sensing images and recurrent information in a single test image, a class-feature capture module (CCM) and an unsupervised learning module (ULM) are leveraged in our framework. Extensive experiments show that our framework outperforms the current state-of-the-art SR algorithms in remotely sensed imagery SR with unknown Gaussian blur kernel.

Real-time video super resolution network using recurrent multi-branch dilated convolutions

Recent developments of video super-resolution reconstruction often exploit spatial and temporal contexts from input frame sequence by making use of explicit motion estimation, e.g., optical flow, which may introduce accumulated errors and requires huge computations to obtain an accurate estimation. In this paper, we propose a novel multi-branch dilated convolution module for real-time frame alignment without explicit motion estimation, which is incorporated with the depthwise separable up-sampling module to formulate a sophisticated real-time video super-resolution network. Specifically, the proposed video super-resolution framework can efficiently acquire a larger receptive field and learn spatial–temporal features of multiple scales with minimal computational operations and memory requirements. Extensive experiments show that the proposed super-resolution network outperforms current state-of-the-art real-time video super-resolution networks, e.g., VESPCN and 3DVSRnet, in terms of PSNR values (0.49 dB and 0.17 dB) on average in various datasets, but requires less multiplication operations.

High-Dimensional Dense Residual Convolutional Neural Network for Light Field Reconstruction.

We consider the problem of high-dimensional light field reconstruction and develop a learning-based framework for spatial and angular super-resolution. Many current approaches either require disparity clues or restore the spatial and angular details separately. Such methods have difficulties with non-Lambertian surfaces or occlusions. In contrast, we formulate light field super-resolution (LFSR) as tensor restoration and develop a learning framework based on a two-stage restoration with 4-dimensional (4D) convolution. This allows our model to learn the features capturing the geometry information encoded in multiple adjacent views. Such geometric features vary near the occlusion regions and indicate the foreground object border. To train a feasible network, we propose a novel normalization operation based on a group of views in the feature maps, design a stage-wise loss function, and develop the multi-range training strategy to further improve the performance. Evaluations are conducted on a number of light field datasets including real-world scenes, synthetic data, and microscope light fields. The proposed method achieves superior performance and less execution time comparing with other state-of-the-art schemes.

Cross-Scale Residual Network: A General Framework for Image Super-Resolution, Denoising, and Deblocking.

In general, image restoration involves mapping from low-quality images to their high-quality counterparts. Such optimal mapping is usually nonlinear and learnable by machine learning. Recently, deep convolutional neural networks have proven promising for such learning processing. It is desirable for an image processing network to support well with three vital tasks, namely: 1) super-resolution; 2) denoising; and 3) deblocking. It is commonly recognized that these tasks have strong correlations, which enable us to design a general framework to support all tasks. In particular, the selection of feature scales is known to significantly impact the performance on these tasks. To this end, we propose the cross-scale residual network to exploit scale-related features among the three tasks. The proposed network can extract spatial features across different scales and establish cross-temporal feature reusage, so as to handle different tasks in a general framework. Our experiments show that the proposed approach outperforms state-of-the-art methods in both quantitative and qualitative evaluations for multiple image restoration tasks.

Real-time video super-resolution using lightweight depthwise separable group convolutions with channel shuffling

In recent years, convolutional neural networks (CNNs) have accelerated the developments of video super resolution (SR) for achieving higher image quality. However, the computational cost of existing CNN-based video super-resolution is too heavy for real-time applications. In this paper, we propose a new video super-resolution framework using lightweight frame alignment module and well-designed up-sampling module for real-time processing. Specifically, our framework, which is called as Lightweight Shuffle Video Super-Resolution Network (LSVSR), combines channel shuffling, depthwise convolution and pointwise group convolution to significantly reduce the computational burden during frame alignment and high-resolution frame reconstruction. On the public benchmark datasets, our proposed network outperforms the state-of-the-art lightweight video SR networks in terms of objective (PSNR and SSIM) and subjective evaluations, number of network parameters and floating-point operations. Our network can achieve real-time 540P to 2160P 4× super-resolution for more than 60fps using desktop GPUs or mobile phones with neural processing unit.

A Transfer Learning Based Super-Resolution Microscopy for Biopsy Slice Images: The Joint Methods Perspective.

Higher-resolution biopsy slice images reveal many details, which are widely used in medical practice. However, taking high-resolution slice images is more costly than taking low-resolution ones. In this paper, we propose a joint framework containing a novel transfer learning strategy and a deep super-resolution framework to generate high-resolution slice images from low-resolution ones. The super-resolution framework called SRFBN+ is proposed by modifying a state-of-the-art framework SRFBN. Specifically, the structure of the feedback block of SRFBN was modified to be more flexible. Besides, it is challenging to use typical transfer learning strategies directly for the tasks on slice images, as the patterns on different types of biopsy slice images are varying. To this end, we propose a novel transfer learning strategy, called Channel Fusion Transfer Learning (CF-Trans). CF-Trans builds a middle domain by fusing the data manifolds of the source domain and the target domain, serving as a springboard for knowledge transfer. Thus, in the transfer learning setting, SRFBN+ can be trained on the source domain and then the middle domain and finally the target domain. Experiments on biopsy slice images validate SRFBN+ works well in generating super-resolution slice images, and CF-Trans is an efficient transfer learning strategy.

Sentinel-3 Super-Resolution Based on Dense Multireceptive Channel Attention

The unprecedented availability of remote sensing data from different complementary Sentinel missions provides increasing opportunities to alleviate the spatial limitations of Sentinel-3 (S3) from an intersensor perspective. Nonetheless, effectively exploiting such intersensor synergies still raises important challenges for super-resolution (SR) algorithms in terms of operational data availability, sensor alignment and substantial resolution changes, among others. In this scenario, this article sets a new SR framework for spatially enhancing S3 ocean and land color instrument (OLCI) products by taking advantage of the higher spatial resolution of the Sentinel-2 (S2) multispectral instrument (MSI). To achieve this goal, we initially study some of the most important deep learning-based approaches. Then, we define a novel Level-4 SR framework which integrates a new convolutional neural network specially designed for super-resolving OLCI data. In contrast to other networks, the proposed SR architecture (termed as SRS3) employs a dense multireceptive field together with a residual channel attention mechanism to relieve the particularly low spatial resolution of OLCI while extracting more discriminating features for the large spatial resolution differences with respect to MSI. The experimental part of the work, conducted using ten coupled OLCI and MSI operational data, reveals the suitability of the presented Level-4 SR framework within the Copernicus programme context as well as the advantages of the proposed architecture with respect different state-of-the-art models when spatially enhancing OLCI products. The related codes will be publicly available at https://github.com/rufernan/SRS3 .

Mars Image Super-Resolution Based on Generative Adversarial Network

High-resolution (HR) Mars images have great significance for studying the land-form features of Mars and analyzing the climate on Mars. Nowadays, the mainstream image super-resolution methods are based on deep learning or CNNs, which are better than traditional methods. However, these deep learning based methods obtain low-resolution(LR) images usually by using an ideal down-sampling method (e.g. bicubic interpolation). There are two limitations in the existing SR methods: 1) The paired LR-HR data by using such methods can achieve a satisfactory results when tested on an ideal datasets. But, these methods always fail in real Mars image super-resolution, since real Mars images rarely obey an ideal down-sampling rule. 2) The LR images obtained by ideal down-sampling methods have no noise while real Mars images usually have noise, which leads to the super-resolved images are not realistic in texture details. To solve the above-mentioned problems, in this article, we propose a novel two-step framework for Mars image super-resolution. Specifically, to address limitation 1), we focus on designing a new degradation framework by estimating blur-kernels. To address limitation 2), a Generative Adversarial Network (GAN) is trained to generate noise distribution. Extensive experiments on the Mars32k dataset demonstrate the effectiveness of the proposed method, and we achieve better qualitative and quantitative results compared to other SOTA methods.

STRESS: Super-Resolution for Dynamic Fetal MRI using Self-Supervised Learning.

Fetal motion is unpredictable and rapid on the scale of conventional MR scan times. Therefore, dynamic fetal MRI, which aims at capturing fetal motion and dynamics of fetal function, is limited to fast imaging techniques with compromises in image quality and resolution. Super-resolution for dynamic fetal MRI is still a challenge, especially when multi-oriented stacks of image slices for oversampling are not available and high temporal resolution for recording the dynamics of the fetus or placenta is desired. Further, fetal motion makes it difficult to acquire high-resolution images for supervised learning methods. To address this problem, in this work, we propose STRESS (Spatio-Temporal Resolution Enhancement with Simulated Scans), a self-supervised super-resolution framework for dynamic fetal MRI with interleaved slice acquisitions. Our proposed method simulates an interleaved slice acquisition along the high-resolution axis on the originally acquired data to generate pairs of low- and high-resolution images. Then, it trains a super-resolution network by exploiting both spatial and temporal correlations in the MR time series, which is used to enhance the resolution of the original data. Evaluations on both simulated and in utero data show that our proposed method outperforms other self-supervised super-resolution methods and improves image quality, which is beneficial to other downstream tasks and evaluations.

Pixel-Level Kernel Estimation for Blind Super-Resolution

Throughout the past several years, deep learning-based models have achieved success in super-resolution (SR). The majority of these works assume that low-resolution (LR) images are ‘uniformly’ degraded from their corresponding high-resolution (HR) images using predefined blur kernels — all regions of an image undergoing an identical degradation process. Furthermore, based on this assumption, there have been attempts to estimate the blur kernel of a given LR image, since correct kernel priors are known to be helpful in super-resolution. Although it has been known that blur kernels of real images are non-uniform (spatially varying), current kernel estimation algorithms are mostly done at image-level, estimating one kernel per image. These algorithms inevitably become sub-optimal in handling scenarios where an image is degraded non-uniformly. A divide-and-conquer form of approach, dividing an image into several patches for individual kernel estimation and SR can be a simple solution for this matter. Nevertheless, this approach fails in practice. In this paper, we address this issue by pixel-level kernel estimation. The three main components for training a SR framework based on pixel-level kernel estimation are as follows: Kernel Collage — a method for synthesizing non-uniformly degraded LR images, designed considering the coherency of kernels at neighboring regions while abruptly changing at times, the indirect loss — a novel loss for training the kernel estimator, based on the reconstruction loss, and an additional optimization — a scheme to robustify the SR network to minor errors in kernel estimations. Extensive experiments show the superiority of pixel-level kernel estimation in blind SR, surpassing state-of-the-art methods in terms of quantitative and qualitative results.

Unsupervised Recurrent Hyperspectral Imagery Super-Resolution Using Pixel-Aware Refinement

Unsupervised fusion-based hyperspectral imagery (HSI) super-resolution (SR) is an essential task of HSI processing, which aims to reconstruct a high-resolution (HR) HSI using only an observed low-resolution HSI and a conventional HR image. Although a large number of unsupervised HSI SR methods have been proposed, the heuristic handcrafted image priors adopted by the majority of these methods restrict their capacity to capture specific characteristics of the HSI, as well as their ability to generalize to noisy observation images. In this study, we investigate a fusion-based HSI SR framework with the deep image prior, in which the deep neural network (rather than a heuristic handcrafted image prior) is exploited to capture plenty of image statistics. Within this framework, we further propose an unsupervised recurrence-based HSI SR method using pixel-aware refinement, which utilizes the intermediate reconstruction results to self-supervise unsupervised learning. Due to containing the information of the image-specific characteristic, the proposed method achieves better performance, in terms of both accuracy and robustness to noise, compared with the existing methods. Extensive experiments on four HSI data sets demonstrate the effectiveness of the proposed method.

Towards Practical Training Framework for Image Super-resolution

Towards Practical Training Framework for Image Super-resolution