Residual Dense Block Research Articles

High-fidelity deconvolution for acoustic-resolution photoacoustic microscopy enabled by convolutional neural networks

Acoustic-resolution photoacoustic microscopy (AR-PAM) image resolution is determined by the point spread function (PSF) of the imaging system. Previous algorithms, including Richardson–Lucy (R–L) deconvolution and model-based (MB) deconvolution, improve spatial resolution by taking advantage of the PSF as prior knowledge. However, these methods encounter the problems of inaccurate deconvolution, meaning the deconvolved feature size and the original one are not consistent (e.g., the former can be smaller than the latter). We present a novel deep convolution neural network (CNN)-based algorithm featuring high-fidelity recovery of multiscale feature size to improve lateral resolution of AR-PAM. The CNN is trained with simulated image pairs of line patterns, which is to mimic blood vessels. To investigate the suitable CNN model structure and elaborate on the effectiveness of CNN methods compared with non-learning methods, we select five different CNN models, while R–L and directional MB methods are also applied for comparison. Besides simulated data, experimental data including tungsten wires, leaf veins, and in vivo blood vessels are also evaluated. A custom-defined metric of relative size error (RSE) is used to quantify the multiscale feature recovery ability of different methods. Compared to other methods, enhanced deep super resolution (EDSR) network and residual in residual dense block network (RRDBNet) model show better recovery in terms of RSE for tungsten wires with diameters ranging from 30 μm to 120 μm. Moreover, AR-PAM images of leaf veins are tested to demonstrate the effectiveness of the optimized CNN methods (by EDSR and RRDBNet) for complex patterns. Finally, in vivo images of mouse ear blood vessels and rat ear blood vessels are acquired and then deconvolved, and the results show that the proposed CNN method (notably RRDBNet) enables accurate deconvolution of multiscale feature size and thus good fidelity.

MFCT-GAN: multi-information network to reconstruct CT volumes for security screening

PurposeAt airport security checkpoints, baggage screening is aimed to prevent transportation of prohibited and potentially dangerous items. Observing the projection images generated by X-rays scanner is a critical method. However, when multiple objects are stacked on top of each other, distinguishing objects only by a two-dimensional picture is difficult, which prompts the demand for more precise imaging technology to be investigated for use. Reconstructing from 2D X-ray images to 3D-computed tomography (CT) volumes is a reliable solution.Design/methodology/approachTo more accurately distinguish the specific contour shape of items when stacked, multi-information fusion network (MFCT-GAN) based on generative adversarial network (GAN) and U-like network (U-NET) is proposed to reconstruct from two biplanar orthogonal X-ray projections into 3D CT volumes. The authors use three modules to enhance the reconstruction qualitative and quantitative effects, compared with the original network. The skip connection modification (SCM) and multi-channels residual dense block (MRDB) enable the network to extract more feature information and learn deeper with high efficiency; the introduction of subjective loss enables the network to focus on the structural similarity (SSIM) of images during training.FindingsOn account of the fusion of multiple information, MFCT-GAN can significantly improve the value of quantitative indexes and distinguish contour explicitly between different targets. In particular, SCM enables features more reasonable and accurate when expanded into three dimensions. The appliance of MRDB can alleviate problem of slow optimization during the late training period, as well as reduce the computational cost. The introduction of subjective loss guides network to retain more high-frequency information, which makes the rendered CT volumes clearer in details.Originality/valueThe authors' proposed MFCT-GAN is able to restore the 3D shapes of different objects greatly based on biplanar projections. This is helpful in security check places, where X-ray images of stacked objects need to be distinguished from the presence of prohibited objects. The authors adopt three new modules, SCM, MRDB and subjective loss, as well as analyze the role the modules play in 3D reconstruction. Results show a significant improvement on the reconstruction both in objective and subjective effects.

NDSRGAN: A Novel Dense Generative Adversarial Network for Real Aerial Imagery Super-Resolution Reconstruction

In recent years, more and more researchers have used deep learning methods for super-resolution reconstruction and have made good progress. However, most of the existing super-resolution reconstruction models generate low-resolution images for training by downsampling high-resolution images through bicubic interpolation, and the models trained from these data have poor reconstruction results on real-world low-resolution images. In the field of unmanned aerial vehicle (UAV) aerial photography, the use of existing super-resolution reconstruction models in reconstructing real-world low-resolution aerial images captured by UAVs is prone to producing some artifacts, texture detail distortion and other problems, due to compression and fusion processing of the aerial images, thereby resulting in serious loss of texture detail in the obtained low-resolution aerial images. To address this problem, this paper proposes a novel dense generative adversarial network for real aerial imagery super-resolution reconstruction (NDSRGAN), and we produce image datasets with paired high- and low-resolution real aerial remote sensing images. In the generative network, we use a multilevel dense network to connect the dense connections in a residual dense block. In the discriminative network, we use a matrix mean discriminator that can discriminate the generated images locally, no longer discriminating the whole input image using a single value but instead in chunks of regions. We also use smoothL1 loss instead of the L1 loss used in most existing super-resolution models, to accelerate the model convergence and reach the global optimum faster. Compared with traditional models, our model can better utilise the feature information in the original image and discriminate the image in patches. A series of experiments is conducted with real aerial imagery datasets, and the results show that our model achieves good performance on quantitative metrics and visual perception.

A Novel Reference-Based and Gradient-Guided Deep Learning Model for Daily Precipitation Downscaling

The spatial resolution of precipitation predicted by general circulation models is too coarse to meet current research and operational needs. Downscaling is one way to provide finer resolution data at local scales. The single-image super-resolution method in the computer vision field has made great strides lately and has been applied in various fields. In this article, we propose a novel reference-based and gradient-guided deep learning model (RBGGM) to downscale daily precipitation considering the discontinuity of precipitation and ill-posed nature of downscaling. Global Precipitation Measurement Mission (GPM) precipitation data, variables in ERA5 re-analysis data, and topographic data are selected to perform the downscaling, and a residual dense attention block is constructed to extract features of them. By exploring the discontinuous feature of precipitation, we introduce gradient feature to reconstruct precipitation distribution. We also extract the feature of high-resolution monthly precipitation as a reference feature to resolve the ill-posed nature of downscaling. Extensive experimental results on benchmark data sets demonstrate that our proposed model performs better than other baseline methods. Furthermore, we construct a daily precipitation downscaling data set based on GPM precipitation data, ERA5 re-analysis data and topographic data.

Fusing Landsat 8 and Sentinel-2 data for 10-m dense time-series imagery using a degradation-term constrained deep network

Dense medium-resolution imagery is essential for fine-scale time-series applications. The combined use of Landsat 8 and Sentinel-2 can derive 10-m time-series imagery at a nominal temporal resolution of ∼ 2.9 days. Specifically, Landsat images can be downscaled to a 10-m resolution by fusing them with temporally adjacent Sentinel-2 images. Current approaches simply use a linear model or a shallow network that is insufficient to obtain the complex mapping between inputs and outputs, and they rarely consider the temporal variation issue, especially for scenes experiencing land cover changes. Facing these limitations, we proposed a degradation-term constrained spatiotemporal fusion network (DSTFN). Technically, a deep network architecture incorporating residual dense blocks and attention mechanism modules is adopted to enhance the feature representation and extraction. A degradation constraint term is embedded into the loss function to maximize the use of the input coarse-resolution image and improve the capability of predicting change. A series of experiments based on two new datasets indicate that DSTFN achieves the best quantitative scores in every test and is thus effective and robust. In 20 resolution-degraded tests, on average, DSTFN decreases the mean relative error by 0.85%–5.35% and increases the peak signal-to-noise ratio 0.97–6.23 relative to baseline approaches. The tests featuring diverse temporal dynamics also confirm the strong generalization ability of DSTFN to deal with land cover change. The proposed network can be used to produce 10-m dense time-series imagery and shows great promise for a variety of time-series analyses and applications. The test materials are expected to be employed as standard datasets for future model assessment.

Neural Style Transfer for image within images and conditional GANs for destylization

In this paper, the feature representation of an image by CNN is used to hide the secret image into the cover image. The style of the cover image hides the content of the secret image and produce a stego image using Neural Style Transfer (NST) algorithm, which resembles the cover image and also contains the semantic content of secret image. The main technical contributions are to hide the content of the secret image in the in-between hidden layered style features of the cover image, which is the first of its kind in the present state-of-art-technique. Also, to recover the secret image from the stego image, destylization is done with the help of conditional generative adversarial networks (GANs) using Residual in Residual Dense Blocks (RRDBs). Further, stego images from different layer combinations of content and style features are obtained and evaluated. Evaluation is based on the visual similarity and quality loss between the cover-stego pair and the secret-reconstructed secret pair of images. From the experiments, it has been observed that the proposed algorithm has 43.95 dB Peak Signal-to-Noise Ratio (PSNR)), .995 Structural Similarity Index (SSIM), and .993 Visual Information Fidelity (VIF) for the ImageNet dataset. The proposed algorithm is found to be more robust against StegExpose than the traditional methods.

Dehazing Based on Long-Range Dependence of Foggy Images

Deep neural networks (DNNs) with long-range dependence (LRD) have attracted more and more attention recently. However, LRD of DNNs is proposed from the view on gradient disappearance in training, which lacks theory analysis. In order to prove LRD of foggy images, the Hurst parameters of over 1,000 foggy images in SOTS are computed and discussed. Then, the Residual Dense Block Group (RDBG), which has additional long skips among two Residual Dense Blocks to fit LRD of foggy images, is proposed. The Residual Dense Block Group can significantly improve the details of dehazing image in dense fog and reduce the artifacts of dehazing image.

A nondestructive automatic defect detection method with pixelwise segmentation

Defect detection is essential for the quality control and repair decision-making of various products. Due to collisions, uneven stress, welding parameters and other factors, cracks form on the surface or inside of products, which affect the product appearance and mechanism strength and may even cause huge safety accidents. Nondestructive testing (NDT) is an effective and practical method for accurate defect detection, but it still faces various challenges against complex factors, such as complex backgrounds, poor contrast, weak texture, and class imbalance issues. Recently, deep learning has rapidly improved the performance of automatic defect detection with the strong feature expression ability of deep convolutional neural networks (DCNNs). However, various limitations remain due to the insufficient processing of local contextual features, which affects the detection precision. To address this issue, with the encoder–decoder network structure, a novel nondestructive defect detection network, namely, NDD-Net, is proposed in this paper to construct an end-to-end nondestructive defect segmentation scheme. To make the segmentation network better emphasize the defect areas, an attention fusion block (AFB) is proposed to replace the raw skip connections to acquire more discriminative features and enhance the segmentation performance on microdefects. Meanwhile, by fusing a dense connection convolution network and a residual network, a residual dense connection convolution block (RDCCB) is also proposed to be embedded into the proposed segmentation network to acquire richer information about the local feature maps. Two public datasets with severe class imbalance issues are adopted for model evaluation: the Grima X-ray (GDXray) database and the rail surface discrete defects (RSSDs) dataset. Experimental results show that the proposed segmentation network outperforms other related segmentation models.

FloodNet: Underwater image restoration based on residual dense learning

The efficiency of the underwater image restoration task is hindered due to the degradation factors like light scattering, color shift, suspended particles and haze occurring in the underwater environment. In recent years, different methods have relied on the underwater image formation model and deep learning techniques to restore the underwater image, but they tend to produce unnatural artifacts and reduced levels of sharpness. To address these challenges, in this paper, we propose FloodNet using residual dense learning with the objective of estimating restored underwater images from a wide variety of degraded underwater images. The proposed FloodNet architecture is a fully convolution neural network that comprises of three modules, namely, low-level feature extraction for extracting features from the degraded underwater image, residual dense blocks that are densely connected via skip-connections for hierarchical feature fusion and to improve the flow of gradient during back-propagation, and finally global feature fusion to adaptively utilize both local and global residual learning to obtain a restored underwater image. We conduct the quantitative and qualitative evaluations on paired and unpaired underwater image datasets, as well as the application and user study analysis. The results demonstrate superior performance of FloodNet against existing SOTA methods.

Downscaling of soil moisture products using deep learning: Comparison and analysis on Tibetan Plateau

Soil moisture (SM) is a very important physical parameter in land surface processes. The low-resolution microwave products are extremely useful for large-scale studies, but are limited in their ability to resolve landscape-level variability. In this study, to overcome this limitation, we adopted three different deep learning based downscaling methods and performed a comparative analysis. Firstly, a downscaling method based on a deep belief network (DBN) model with a fully connected structure was employed to fit the complex nonlinear relationship between soil moisture and the auxiliary surface parameters. Then, to improve this simple point-to-point relationship, a neighborhood constraint based improved DBN model was developed to consider the effect of local spatial correlation among the surface factors on soil moisture. Finally, to maintain the integrity of the spatial structure in the process of model construction, a residual network (ResNet) model consisting of several residual dense blocks (RDBs) was introduced to use convolutions to obtain more complete and powerful spatial feature maps. The Tibetan Plateau was selected as the study area because of its importance to the global water cycle. The four soil moisture monitoring networks in the research area were used to verify the effectiveness of the proposed methods in the downscaling of soil moisture products, and four quantitative indicators were used to evaluate the effectiveness of the methods. From the verified results, it is shown that the three deep learning methods can maintain a higher correlation and better reduce the bias, on the whole, and the ResNet model, in particular, shows a higher stability. From the perspective of the spatial patterns and details, when compared to back-propagation neural network (BPNN) and random forest (RF) models, the three deep learning methods can provide more feasible spatial patterns and details, as well as reducing the uncertainty in the central regions of the Tibetan Plateau. The ResNet model shows a superior ability to enhance the texture details with lower uncertainty, due to the fact that it can meet the requirements of fitting the soil moisture to complex surfaces. Furthermore, the results of the ResNet model can also effectively capture the temporal changes of the microwave soil moisture product well.

Integration of dilated convolution with residual dense block network and multi‐level feature detection network for cassava plant leaf disease identification

AbstractIn recent years, many countries prolonged their research in the field of agricultural farming for yielding more quantity food products with good quality. Agricultural farming is closely related to every country's economy in significant way. To improve the grain productivity in a notable way, plants must be protected from various viral, fungal, and bacterial diseases. The timely identification of plant disease helps to protect the plant at the early stage of the disease else it will spoil the entire plant. In this article, the plant disease identification problem is handled using our novel proposed network where it combines dilated convolution with residual dense block (DCRDB) along with multi‐level feature detection (MLFD) for selecting the appropriate feature and bidirectional long short‐term memory (Bi‐LSTM) classifier for leaf disease prediction. In DCRDB, the dilated convolution derives larger receptive field without loss of resolution to extract greater number of local features from the leaf images and residual dense block expands the size of feature map by densely connecting the layers of convolution. In addition, MLFD network integrates the high‐level features with the resulted features of DCRDB. After feature extraction, Bi‐LSTM is employed to recognize the leaf disease which supports additional training in both forward and backward direction. Through a set of collected cassava leaf images our research carries forward. The proposed model accuracy is verified with three different datasets such as cassava disease classification dataset, self‐made cassava disease dataset and plant village dataset. The performance of the proposed model is compared with some existing methods. Experimental results show that the highest F1‐score value of 95.49% attained by our proposed model for identifying cassava leaf disease.

Multiscale Supervision-Guided Context Aggregation Network for Single Image Dehazing

End-to-end learning-based image dehazing methods tend to overdehaze or underdehaze in real scenes due to inefficient feature extraction and feature fusion. In this letter, we propose a multiscale supervision-guided context aggregation network (MSGCAN) based on two principles: improving feature extraction and enhancing feature mapping. To improve feature extraction, an attention-guided context aggregation (AGCA) module is adopted to merge context features extracted by several residual dense blocks (RDB). Moreover, we output these aggregated context features on each scale and form multiscale supervision to enhance feature mapping and ensure that the extracted features on each scale contain more realistic details. The experimental results show that the proposed MSGCAN performs better than other state-of-the-art dehazing methods in both synthetic and real-world scenes.

Image Super Resolution using Enhanced Super Resolution Generative Adversarial Network

Aside from enhancing the accuracy and speed of single picture modification utilizing fast and in-depth convolutional emotional networks, one significant challenge remains mostly commonly unaddressed, namely how do we recover soft texture details when we concentrate too much on exceptional improvement features? The resultant evaluations offer greater transmission ratings, but the high frequency data is non-existent and unsatisfactory mostly in sense that now it fails to meet the consistency anticipated in high resolution. The resulting ratings have higher signal-to-audio ratings, but the high frequency data is non-existent and unsatisfactory in the sense that it fails to match the consistency expected in high resolution. Introducing ESRGAN, an Advanced Optical Genetically Modified (GAN) network of high-resolution image (SR). To our knowledge, it is a framework capable of identifying immature real-world images up to 4x points rising. To achieve this, we propose a function of loss of vision that combines the loss of content with the loss of content (Mean Squared Error Loss). Controversial Loss has our solution for many uncooked pictures utilizing a discriminatory network which is taught to distinguish between high resolution images and realistic images. We have built a structure that contains several RRDB blocks (Residual in Residual Dense Block) outside the Batch Normalization layers. Our deep residual network can find realistic image texture in very low sample images. Additionally, we used techniques that included residual measurement and a small implementation to train a deeper model. We also introduced the relativistic GAN as a racist, who learns to judge whether one image is more realistic than another, directing the generator to return a detailed texture. In addition, we have improved vision loss by using features before activation, which provides greater security and thus restoring more precise light and texture.

SWCGAN: Generative Adversarial Network Combining Swin Transformer and CNN for Remote Sensing Image Super-Resolution

Easy and efficient acquisition of high-resolution remote sensing images is of importance in geographic information systems. Previously, deep neural networks composed of convolutional layers have achieved impressive progress in super-resolution reconstruction. However, the inherent problems of the convolutional layer, including the difficulty of modeling the long-range dependency, limit the performance of these networks on super-resolution reconstruction. To address the above problems, we propose a generative adversarial network (GAN) by combining the advantages of the swin transformer and convolutional layers, called SWCGAN. It is different from the previous super-resolution models, which are composed of pure convolutional blocks. The essential idea behind the proposed method is to generate high-resolution images by a generator network with a hybrid of convolutional and swin transformer layers and then to use a pure swin transformer discriminator network for adversarial training. In the proposed method, (1) we employ a convolutional layer for shallow feature extraction that can be adapted to flexible input sizes; (2) we further propose the residual dense swin transformer block (RDSTB) to extract deep features for upsampling to generate high-resolution images; and (3) we use a simplified swin transformer as the discriminator for adversarial training. To evaluate the performance of the proposed method, we compare the proposed method with other state-of-the-art methods by utilizing the UCMerced benchmark dataset, and we apply the proposed method to real-world remote sensing images. The results demonstrate that the reconstruction performance of the proposed method outperforms other state-of-the-art methods in most metrics

Image fusion in the loop of high-level vision tasks: A semantic-aware real-time infrared and visible image fusion network

Infrared and visible image fusion aims to synthesize a single fused image that not only contains salient targets and abundant texture details but also facilitates high-level vision tasks. However, the existing fusion algorithms unilaterally focus on the visual quality and statistical metrics of fused images but ignore the demands of high-level vision tasks. To address these challenges, this paper bridges the gap between image fusion and high-level vision tasks and proposes a semantic-aware real-time image fusion network (SeAFusion). On the one hand, we cascade the image fusion module and semantic segmentation module and leverage the semantic loss to guide high-level semantic information to flow back to the image fusion module, which effectively boosts the performance of high-level vision tasks on fused images. On the other hand, we design a gradient residual dense block (GRDB) to enhance the description ability of the fusion network for fine-grained spatial details. Extensive comparative and generalization experiments demonstrate the superiority of our SeAFusion over state-of-the-art alternatives in terms of maintaining pixel intensity distribution and preserving texture detail. More importantly, the performance comparison of various fusion algorithms in task-driven evaluation reveals the natural advantages of our framework in facilitating high-level vision tasks. In addition, the superior running efficiency allows our algorithm to be effortlessly deployed as a real-time pre-processing module for high-level vision tasks. The source code will be released at https://github.com/Linfeng-Tang/SeAFusion.

Infrared Small UAV Target Detection Based on Depthwise Separable Residual Dense Network and Multiscale Feature Fusion

Unmanned aerial vehicles (UAVs) have been widely applied in military and civilian fields, but they also pose great threats to restricted areas, such as densely populated areas and airports. Thermal infrared imaging technology is capable of monitoring UAVs at a long range in both day and night conditions. Therefore, the anti-UAV technology based on thermal infrared imaging has attracted growing attention. However, the images acquired by infrared sensors often suffer from small and dim targets as well as heavy background clutter and noise. Conventional detection methods usually have a high false alarm rate and low detection accuracy. This paper proposes a detection method that formulates the UAV detection as predicting the residual image (i.e., background, clutter, and noise) by learning the nonlinear mapping from the input image to the residual image. The UAV target image is obtained by subtracting the residual image from the input infrared image. The constructed end-to-end U-shaped network exploits the depthwise separable residual dense blocks in the encoder stage to extract the abundant hierarchical features. Besides, the multi-scale feature fusion and representation block is introduced to fully aggregate multi-scale features from the encoder layers and intermediate connection layers at the same scale as well as the decoder layers at different scales to better reconstruct the residual image in the decoder stage. In addition, the global residual connection is adopted in the proposed network to provide long-distance information compensation and promote gradient backpropagation, which further improves the performance in reconstructing the image. The experimental results show that the proposed method achieves favorable detection performance in real-world infrared images and outperforms other state-of-the-art methods in terms of quantitative and qualitative evaluation metrics.

Multistage Dual-Attention Guided Fusion Network for Hyperspectral Pansharpening

Deep learning, especially the convolutional neural network, has been widely applied to solve the hyperspectral pansharpening problem. However, most do not explore the intraimage characteristics and the interimage correlation concurrently due to the limited representation ability of the networks, which may lead to insufficient fusion of valuable information encoded in the high-resolution panchromatic images (HR-PANs) and low-resolution hyperspectral images (LR-HSIs). To cope with this problem, we develop a hyperspectral pansharpening method called multistage dual-attention guided fusion network (MDA-Net) to fully extract the important information and accurately fuse them. It employs a three-stream structure, which enables the network to incorporate the intrinsic characteristics of each input and correlation among them simultaneously. In order to combine as much information as possible, we merge the features extracted from three streams in multiple stages, where a dual-attention guided fusion block (DAFB) with spectral and spatial attention mechanisms is utilized to fuse the features efficiently. It identifies the useful components in both spatial and spectral domains, which are beneficial to improving the fusion accuracy. Moreover, we design a multiscale residual dense block (MRDB) to extract dense and hierarchical features, which improves the representation power of the network. Experiments are conducted on both real and simulated datasets. The evaluation results validate the superiority of the MDA-Net.

Learning to Remove Clutter in Real-World GPR Images Using Hybrid Data

The clutter in the ground-penetrating radar (GPR) radargram disguises or distorts subsurface target responses, which severely affects the accuracy of target detection and identification. Existing clutter removal methods either leave residual clutter or deform target responses when facing complex and irregular clutter in the real-world radargram. To tackle the challenge of clutter removal in real scenarios, a clutter-removal neural network (CR-Net) trained on a large-scale hybrid dataset is presented in this study. The CR-Net integrates residual dense blocks into the U-Net architecture to enhance its capability in clutter suppression and target reflection restoration. The combination of the mean absolute error (MAE) loss and the multi-scale structural similarity (MS-SSIM) loss is used to effectively drive the optimization of the network. To train the proposed CR-Net to remove complex and diverse clutter in real-world radargrams, the first large-scale hybrid dataset named CLT-GPR dataset containing clutter collected by different GPR systems in multiple scenarios is built. The CLT-GPR dataset significantly improves the generalizability of the network to remove clutter in real-world GPR radargrams. Extensive experimental results demonstrate that the CR-Net achieves superior performance over existing methods in removing clutter and restoring target responses in diverse real-world scenarios. Moreover, the CR-Net with its end-to-end design does not require manual parameter tuning, making it highly suitable for automatically producing clutter-free radargrams in GPR applications. The CLT-GPR dataset and the code implemented in the paper can be found at https://haihan-sun.github.io/GPR.html.

Deep-ASL enhancement technique in arterial spin labeling MRI – A novel approach for the error reduction of partial volume correction technique with linear regression algorithm

Arterial Spin Labelling MRI is a noninvasive quantitative imaging technique for measuring Cerebral Blood Flow (CBF) that plays a vital role in diagnosing different neurological disorders. Limited signal-to-noise ratio and significant partial volume effect due to the low resolution of ASL images make an accurate CBF estimation difficult. This work proposes a deep learning based ASL enhancement algorithm (Deep-ASL ENHANCE), based on the principle of single image super resolution and multi-loss joint strategy with two reconstruction modules and one weighted fusion module that employ residual dense block as the basic building block. Lack of huge amount of low quality and high quality images for training this deep learning network, is addressed by generating simulated ASL images from structural images of ADNI2. The experiment is conducted and results are evaluated on a simulated dataset in terms of different metrics such as RMSE, PSNR and SSIM. The model is also validated using clinical ASL images with the help of two independent radiologists and the results are compared using Visual Quality Score (VQS). The deep learning model trained by using simulated ASL images shows more promising results on clinical ASL data. The effectiveness of using Deep-ASL ENHANCE as a preprocessing step to the partial volume correction technique with Linear Regression algorithm (LR) has been investigated using RMSE score and it is found that CBF quantification accuracy is improved compared to the standalone LR algorithm.

Contextual Information Aided Generative Adversarial Network for Low-Light Image Enhancement

Low-light image enhancement has been gradually becoming a hot research topic in recent years due to its wide usage as an important pre-processing step in computer vision tasks. Although numerous methods have achieved promising results, some of them still generate results with detail loss and local distortion. In this paper, we propose an improved generative adversarial network based on contextual information. Specifically, residual dense blocks are adopted in the generator to promote hierarchical feature interaction across multiple layers and enhance features at multiple depths in the network. Then, an attention module integrating multi-scale contextual information is introduced to refine and highlight discriminative features. A hybrid loss function containing perceptual and color component is utilized in the training phase to ensure the overall visual quality. Qualitative and quantitative experimental results on several benchmark datasets demonstrate that our model achieves relatively good results and has good generalization capacity compared to other state-of-the-art low-light enhancement algorithms.