Search

We propose an image super-resolution method (SR) using a deeply-recursive convolutional network (DRCN). Single-image super-resolution reconstruction technology is to reconstruct fuzzy low-resolution images into clearer high-resolution images. It is a research hotspot in the field of computer vision and image processing. In recent years, the attention mechanism has been successfully applied in image super-resolution reconstruction. However, the existing methods use the channel attention mechanism and the spatial attention mechanism separately, or simply superimpose them, which cannot effectively unify the adjustment effects of both, and the performance is limited. This paper proposes a method that can merge channel attention and spatial attention into pixel attention, which achieves more precise adjustment of feature map information. The pixel attention convolutional neural network method built on this basis can improve the quality of image texture detail reconstruction. We have been tested on five widely used standard datasets, the experimental results show that the method is superior to most current representative reconstruction methods, especially in terms of high-definition picture texture restoration. [ABSTRACT FROM AUTHOR]

With the development of image feature matching technology, feature matching algorithms based on deep learning have achieved excellent results, but in scenarios with low texture or extreme perspective changes, the matching accuracy is still difficult to guarantee. In this paper, a superresolution reconstruction method based on a Residual-ESPCN (efficient subpixel convolutional neural network) approach is proposed based on LoFTR (local feature matching with transformers). The superresolution method is used to improve the interpolation method used in ASFF (adaptive spatial feature fusion) to increase the image resolution, enhance the detailed information of the image, and make the extracted features richer. Then, ASFF is introduced into the local feature extraction module of LoFTR, which can alleviate the inconsistency problem of information transmission between different scale features of the feature pyramid and lessen the amount of information lost during transmission from low- to high-resolution levels. Moreover, to improve the adaptability of the algorithm to different scenarios, OTSU is introduced to adaptively calculate the threshold of feature matching. The experimental results show that in different indoor or outdoor scenarios, our proposed algorithm for matching features can effectively improve the adaptability of feature matching and can achieve good results in terms of the area under the curve (AUC), accuracy and recall. [ABSTRACT FROM AUTHOR]

Recently, convolutional neural networks (CNNs) have facilitated the rapid development of image super-resolution. Most deep networks are challenging to apply to the real world due to their high cost of memory storage and computational complexity. This paper addresses this issue by proposing a lightweight subpixel sampling network (SSN). Specifically, we use a traditional encoder-decoder structure and replace the deconvolution and pooling layers by subpixel up-sampling and down-sampling without parameters. Subpixel sampling retains more image information than other sampling methods. In addition, we propose parsimonious spatial and channel attention blocks through which multi-scale features are fused and more image textures can be recovered. Through extensive experiments, we validate the effectiveness of subpixel sampling, spatial attention block, and channel attention block. In terms of quantitative metrics and visual quality, our models achieve performance comparable to state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Deep convolutional neural networks can effectively improve the performance of single-image super-resolution reconstruction. Deep networks tend to achieve better performance than others. However, the deep CNNs will lead to a dramatic increase in the size of parameters, limiting its application on embedding and resource-constrained devices, such as smart phone. To address the common problems of blurred image edges, inflexible convolution kernel size selection and slow convergence during training procedure due to redundant network structure in image super-resolution algorithms, this paper proposes a lightweight single-image super-resolution network that fuses multi-level features. The components are mainly two-level nested residual blocks. To better extract features and reduce the number of parameters, each residual block adopts an asymmetric structure. Firstly, it expands twice and then compresses the number of channels twice. Secondly, in the residual block, the feature information of different channels is weighted and fused by adding an autocorrelation weight unit. The quality of the reconstructed image of the proposed method is superior to the existing image super-resolution reconstruction methods in both subjective perception and objective evaluation indicators, and the reconstruction performance is better when the factor is large. [ABSTRACT FROM AUTHOR]

Super-resolution has achieved remarkable results in recent years, which is attributed to the rapid development of convolutional neural networks (CNN). However, most CNN-based algorithms have complex network structures and are not very adaptive in reconstructing detailed information. In this paper, we propose an adaptive global cross-scale prior attention network (GCPAN) to address this problem. GCPAN consists of nonlocal similarity prior module(NLS), adaptive cross-scale prior fusion module (CSF), and adaptive wide activation residual attention module (WARA). Specifically, we propose the nonlocal similarity prior module (NLS) to extract contextual associations between feature layers to alleviate the pressure of network complexity. NLS ensures the accuracy of the obtained global information utilizing SENet attention. Secondly, the adaptive cross-scale prior fusion (CSF) module is proposed to capture similarity relations of different scales adaptively and takes advantage of the internal prior information of the image to enhance texture details. Finally, we design an adaptive wide activation residual attention (WARA) module to learn more effective detail information. Extensive ablation and comparative experiments show that GCPAN obtains pleasing objective values and is visually closer to real images compared with other state-of-the-art methods. [ABSTRACT FROM AUTHOR]

The main aim of Super Resolution (SR) is to obtain high quality image with high resolution from Low Resolution (LR) images. In numerous computer vision systems, the SR from a single image plays avital role.A major problem in Image SR reconstruction is to acquire a high-resolution (HR) image from a single LR image or multiple LR images which are having sub-pixel changes. Currently, learning based methodology is developed from the HR images from the LR images with better quality. This work presented a spatial entropy based normalization dependent deep convolutional neural network (SNDCN) for SR reconstruction with minimized artifact. Initially, the input LR images are pre-processed utilizing adaptive weighted linear smoothing (AWLS) filtering. Then the pre-processed smoothened images are undergoes to adaptive real time dual threshold (ARDT) technique for preserving the edges. Here, the edge preservation process utilizing the ARDT to enhance the quality of the image. Finally, SNDCN is utilized to construct SR images. The experimental outcomes of the presented technique proved that the varying performance measures attaining better super resolution image reconstruction than the existing methodologies. Additionally, the proposed method complete the reconstruction procedure within 18 s and obtain 34 dB as PSNR value for scaling factor 4 in which the outcome shows that significant of proposed methodology over existing methodologies. [ABSTRACT FROM AUTHOR]

In recent years, convolutional neural networks have obtained significant success in super-resolution (SR) with remarkable performance. However, as the depth of the network increases, the model parameters and computational overhead become complex. In this paper, we propose a lightweight adaptive enhanced attention network named LAEAN by cascading a series of shared adaptive enhanced attention modules (AEAM). Specifically, the AEAM is implemented by adding a novel up-down sampling module (UDSM) in the position and channel attention modules (PCAM) to adaptively capture wider and richer contextual information. The UDSM is designed to aggregate representative features while enabling the PCAM to learn more discriminative features. Furthermore, we propose a weighted fusion module (WFM) to flexibly combine informative features from all AEAMs for further boosting reconstruction performance. The experimental results demonstrate that our LAEAN is efficient and lightweight with only ∼ 1.5M training parameters, and outperforms most state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Recent advances in image super-resolution (SR) explore the power of deep learning to achieve a better reconstruction performance. The deep reconstruction methods mostly reconstruct the super-resolution image based on a large amount of external data training. In order to make full use of the internal features of the image, we propose a novel network for SR called super-resolution network by fusing internal and external features (SRNIF). Specifically, we use the self-similar image itself as the prior knowledge and propose a fully convolutional deep neural network (called Internal Feature Enhancement module, IFE) to extract the internal features of the image so as to obtain detailed texture of the inputted image. Then, the detailed features extracted by the internal feature enhancement module are embedded into the residual network, and the spatial feature transform (SFT) module is used to form the residual module. A deep residual network based on the fusion of internal and external features is built. The proposed method comes with a strong early reconstruction ability and can create the final high-resolution image step by step. Experimental results show that the method has enhanced the texture characteristics and quality of the reconstructed image, and the visual effect and peak signal-to-noise ratio (PSNR) has been improved. [ABSTRACT FROM AUTHOR]

As an important display mode of underwater environments, the sonar image has limitations on the resolution, which often leads to problems with low resolution of underwater objects. Therefore, the image super-resolution algorithm is needed to transform the images from low-resolution to high-resolution. It can improve the visual effect and contribute to subsequent processing such as 3D reconstruction and object recognition. This paper proposes a method for sonar image super-resolution based on generative adversarial networks (GAN). By comparing the super-resolution effects of various interpolation and convolutional neural network algorithms on sonar images, a Residual-in-Residual Dense Block network is employed as the generator of GAN since it has the low distortion and high perceptual quality. Because the sonar image training set does not have enough data, the generator utilizes the transfer learning on the sonar images to produce an optimized network model which is more suitable for super-resolution of sonar image. The VGG19 network is employed as the discriminator. In addition, the perceptual loss is introduced into the loss function of S 2 RGAN to further improve the perceptual quality of super-resolution images. The experimental results indicate that the proposed S 2 RGAN shows excellent performance. The generated super-resolution images of S 2 RGAN have the remarkable advantages of both lower distortion and higher perceptual quality comparing with other methods. Because S 2 RGAN focuses more on the reality and overall visual effect of super-resolution sonar images, it is suitable for various underwater situations. [ABSTRACT FROM AUTHOR]

Due to the limitations of infrared imaging principles and imaging systems, many problems are typically encountered with collected infrared images, such as low resolution, insufficient detail information, and blurred edges. In response to these problems, a method of infrared image super-resolution reconstruction that uses recursive attention and is based on a generative adversarial network is proposed. First, according to the characteristics of low-resolution infrared images such as uniform pixel distributions, low contrast, and poor perceived quality, a deep generator structure with a recursive-attention network is designed in this article. The recursive-attention module is used to extract high-frequency information from the feature maps, suppress useless information, and enhance the expressiveness of the features, which facilitates the reconstruction of texture details of infrared images. Then, to better distinguish the reconstructed images from the original high-resolution images, we designed a discriminator that was composed of a deep convolutional neural network. In addition, targeted improvements were made to the content loss function of GAN. We used the pre-trained VGG-19 network features before activation to calculate the perceptual loss, which helps recover the texture details of the infrared images. The experimental results on infrared image datasets demonstrated that the reconstruction performance of the proposed method is higher than those of several typical methods, and it realizes higher image visual quality. [ABSTRACT FROM AUTHOR]

Image Super-Resolution is a complex method capable of converting a low-resolution image to a high-resolution image. Regarding many challenges of the Super-Resolution problem and a wide variety of its applications in image processing and man's interpretations, it is very crucial to find an operational method. Development of deep learning methods, especially Convolutional Neural Networks has increased the power of image enhancement methods including image Super-Resolution. The goal of this research is to propose a fast speed image Super-Resolution method using Convolutional Neural Networks. The proposed DetailNet network has a small structure to prevent the problems of training very deep networks. Super-Resolution is fast by this method due to its small simple network structure. The DetailNet network is designed to add details to an input image. According to the results, DetailNet has a significant ability to produce image details. This network has a general function. Therefore, the low-resolution image size can be increased first using any method, then DetailNet can enhance the image quality and add more details to the input image. The Proposed method is applied to natural color images which achieved acceptable results on benchmark datasets. [ABSTRACT FROM AUTHOR]