XSIM: A structural similarity index measure optimized for MRI QSM.

PURPOSE: The structural similarity index measure (SSIM) has become a popular quality metric to evaluate QSM in a way that is closer to human perception than RMS error (RMSE). However, SSIM may overpenalize errors in diamagnetic tissues and underpenalize them in paramagnetic tissues, resulting in biasing. In addition, extreme artifacts may compress the dynamic range, resulting in unrealistically high SSIM scores (hacking). To overcome biasing and hacking, we propose XSIM: SSIM implemented in the native QSM range, and with internal parameters optimized for QSM. METHODS: We used forward simulations from a COSMOS ground-truth brain susceptibility map included in the 2016 QSM Reconstruction Challenge to investigate the effect of QSM reconstruction errors on the SSIM, XSIM, and RMSE metrics. We also used these metrics to optimize QSM reconstructions of the in vivo challenge data set. We repeated this experiment with the QSM abdominal phantom. To validate the use of XSIM instead of SSIM for QSM quality assessment across a range of different reconstruction techniques/algorithms, we analyzed the reconstructions submitted to the 2019 QSM Reconstruction Challenge 2.0. RESULTS: Our experiments confirmed the biasing and hacking effects on the SSIM metric applied to QSM. The XSIM metric was robust to those effects, penalizing the presence of streaking artifacts and reconstruction errors. Using XSIM to optimize QSM reconstruction regularization weights returned less overregularization than SSIM and RMSE. CONCLUSION: XSIM is recommended over traditional SSIM to evaluate QSM reconstructions against a known ground truth, as it avoids biasing and hacking effects and provides a larger dynamic range of scores.

Video dataset containing video quality assessment scores obtained from standardized objective and subjective testing.

This paper presents a dataset comprising 700 video sequences encoded in the two most popular video formats (codecs) of today, H.264 and H.265 (HEVC). Six reference sequences were encoded under different quality profiles, including several bitrates and resolutions, and were affected by various packet loss rates. Subsequently, the image quality of encoded video sequences was assessed by subjective, as well as objective, evaluation. Therefore, the enclosed spreadsheet contains results of both assessment approaches in a form of MOS (Mean Opinion Score) delivered by the absolute category ranking (ACR) procedure, SSIM (Structural Similarity Index Measure) and VMAF (Video Multimethod Assessment Fusion). All assessments are available for each test sequence. This allows a comprehensive evaluation of coding efficiency under different test scenarios without the necessity of real observers or a secure laboratory environment, as recommended by the ITU (International Telecommunication Union). As there is currently no standardized mapping function between the results of subjective and objective methods, this dataset can also be used to design and verify experimental machine learning algorithms that contribute to solving the relevant research issues.

SM-CycleGAN: crop image data enhancement method based on self-attention mechanism CycleGAN.

Crop disease detection and crop baking stage judgement require large image data to improve accuracy. However, the existing crop disease image datasets have high asymmetry, and the poor baking environment leads to image acquisition difficulties and colour distortion. Therefore, we explore the potential of the self-attention mechanism on crop image datasets and propose an innovative crop image data-enhancement method for recurrent generative adversarial networks (GANs) fused with the self-attention mechanism to significantly enhance the perception and information capture capabilities of recurrent GANs. By introducing the self-attention mechanism module, the cycle-consistent GAN (CycleGAN) is more adept at capturing the internal correlations and dependencies of image data, thus more effectively capturing the critical information among image data. Furthermore, we propose a new enhanced loss function for crop image data to optimise the model performance and meet specific task requirements. We further investigate crop image data enhancement in different contexts to validate the performance and stability of the model. The experimental results show that, the peak signal-to-noise ratio of the SM-CycleGAN for tobacco images and tea leaf disease images are improved by 2.13% and 3.55%, and the structural similarity index measure is improved by 1.16% and 2.48% compared to CycleGAN, respectively.

Enhanced Spatial Fuzzy C-Means Algorithm for Brain Tissue Segmentation in T1 Images.

Magnetic Resonance Imaging (MRI) plays an important role in neurology, particularly in the precise segmentation of brain tissues. Accurate segmentation is crucial for diagnosing brain injuries and neurodegenerative conditions. We introduce an Enhanced Spatial Fuzzy C-means (esFCM) algorithm for 3D T1 MRI segmentation to three tissues, i.e. White Matter (WM), Gray Matter (GM), and Cerebrospinal Fluid (CSF). The esFCM employs a weighted least square algorithm utilizing the Structural Similarity Index (SSIM) for polynomial bias field correction. It also takes advantage of the information from the membership function of the last iteration to compute neighborhood impact. This strategic refinement enhances the algorithm's adaptability to complex image structures, effectively addressing challenges such as intensity irregularities and contributing to heightened segmentation accuracy. We compare the segmentation accuracy of esFCM against four variants of FCM, Gaussian Mixture Model (GMM) and FSL and ANTs algorithms using four various dataset, employing three measurement criteria. Comparative assessments underscore esFCM's superior performance, particularly in scenarios involving added noise and bias fields.The obtained results emphasize the significant potential of the proposed method in the segmentation of MRI images.

Blockchain based medical image encryption using Arnold's cat map in a cloud environment.

Improved software for processing medical images has inspired tremendous interest in modern medicine in recent years. Modern healthcare equipment generates huge amounts of data, such as scanned medical images and computerized patient information, which must be secured for future use. Diversity in the healthcare industry, namely in the form of medical data, is one of the largest challenges for researchers. Cloud environment and the Block chain technology have both demonstrated their own use. The purpose of this study is to combine both technologies for safe and secure transaction. Storing or sending medical data through public clouds exposes information into potential eavesdropping, data breaches and unauthorized access. Encrypting data before transmission is crucial to mitigate these security risks. As a result, a Blockchain based Chaotic Arnold's cat map Encryption Scheme (BCAES) is proposed in this paper. The BCAES first encrypts the image using Arnold's cat map encryption scheme and then sends the encrypted image into Cloud Server and stores the signed document of plain image into blockchain. As blockchain is often considered more secure due to its distributed nature and consensus mechanism, data receiver will ensure data integrity and authenticity of image after decryption using signed document stored into the blockchain. Various analysis techniques have been used to examine the proposed scheme. The results of analysis like key sensitivity analysis, key space analysis, Information Entropy, histogram correlation of adjacent pixels, Number of Pixel Change Rate, Peak Signal Noise Ratio, Unified Average Changing Intensity, and similarity analysis like Mean Square Error, and Structural Similarity Index Measure illustrated that our proposed scheme is an efficient encryption scheme as compared to some recent literature. Our current achievements surpass all previous endeavors, setting a new standard of excellence.

Breaking boundaries: TINTO in POKY for computer vision-based NMR walking strategies.

Nuclear magnetic resonance is a crucial technique for studying biological complexes, as it provides precise structural and dynamic information at the atomic level. However, the process of assigning resonances can be time-consuming and challenging, particularly in cases where peaks overlap, or the data quality is poor. In this paper, we present TINTO (Two and three-dimensional Imaging for NMR sTrip Operation via CV/ML), an advanced semiautomatic toolset for NMR resonance assignment. TINTO comprises two separate tools, each tailored for either two-dimensional or three-dimensional imaging. The toolset utilizes a computer-vision approach and a machine learning approach, specifically structural similarity index and principal components analysis, to perform visual similarity searches of resonances and quickly locate similar strips, and in that way overcome the challenges associated with peak overlap without requiring peak picking. Our tool offers a user-friendly interface and has the potential to enhance the efficiency and accuracy of NMR resonance assignment, particularly in complex cases. This advancement holds promising implications for furthering our understanding of biological systems at the molecular level. TINTO is pre-installed in the POKY suite, which is available at https://poky.clas.ucdenver.edu .

An Improved BM3D Algorithm Based on Image Depth Feature Map and Structural Similarity Block-Matching.

We propose an improved BM3D algorithm for block-matching based on UNet denoising network feature maps and structural similarity (SSIM). In response to the traditional BM3D algorithm that directly performs block-matching on a noisy image, without considering the deep-level features of the image, we propose a method that performs block-matching on the feature maps of the noisy image. In this method, we perform block-matching on multiple depth feature maps of a noisy image, and then determine the positions of the corresponding similar blocks in the noisy image based on the block-matching results, to obtain the set of similar blocks that take into account the deep-level features of the noisy image. In addition, we improve the similarity measure criterion for block-matching based on the Structural Similarity Index, which takes into account the pixel-by-pixel value differences in the image blocks while fully considering the structure, brightness, and contrast information of the image blocks. To verify the effectiveness of the proposed method, we conduct extensive comparative experiments. The experimental results demonstrate that the proposed method not only effectively enhances the denoising performance of the image, but also preserves the detailed features of the image and improves the visual quality of the denoised image.

Metabolite Annotation Confidence Score (MACS): A Novel MSI Identification Scoring Tool.

Mass spectrometry imaging (MSI) is an analytical technique capable of measuring and visualizing the spatial distribution of thousands of ions across a sample. Measured ions can be putatively identified and annotated by comparing their mass-to-charge ratio (m/z) to a database of known compounds. For high-resolution, accurate mass (HRAM) imaging data sets, this is commonly performed by the annotation platform METASPACE. Annotations are reported with a metabolite-signal-match (MSM) score as a measure of the annotation's confidence level. However, the MSM scores reported by METASPACE often do not reflect a reasonable confidence level of an annotation and are not assigned consistently. The metabolite annotation confidence score (MACS) is an alternative scoring system based on fundamental mass spectrometry imaging metrics (mass measurement accuracy, spectral accuracy, and spatial distribution) to generate values that reflect the confidence of a specific annotation in HRAM-MSI data sets. Herein, the MACS system is characterized and compared to MSM scores from ions annotated by METASPACE.

Subjective Quality Assessment of V-PCC-Compressed Dynamic Point Clouds Degraded by Packet Losses.

This article describes an empirical exploration on the effect of information loss affecting compressed representations of dynamic point clouds on the subjective quality of the reconstructed point clouds. The study involved compressing a set of test dynamic point clouds using the MPEG V-PCC (Video-based Point Cloud Compression) codec at 5 different levels of compression and applying simulated packet losses with three packet loss rates (0.5%, 1% and 2%) to the V-PCC sub-bitstreams prior to decoding and reconstructing the dynamic point clouds. The recovered dynamic point clouds qualities were then assessed by human observers in experiments conducted at two research laboratories in Croatia and Portugal, to collect MOS (Mean Opinion Score) values. These scores were subject to a set of statistical analyses to measure the degree of correlation of the data from the two laboratories, as well as the degree of correlation between the MOS values and a selection of objective quality measures, while taking into account compression level and packet loss rates. The subjective quality measures considered, all of the full-reference type, included point cloud specific measures, as well as others adapted from image and video quality measures. In the case of image-based quality measures, FSIM (Feature Similarity index), MSE (Mean Squared Error), and SSIM (Structural Similarity index) yielded the highest correlation with subjective scores in both laboratories, while PCQM (Point Cloud Quality Metric) showed the highest correlation among all point cloud-specific objective measures. The study showed that even 0.5% packet loss rates reduce the decoded point clouds subjective quality by more than 1 to 1.5 MOS scale units, pointing out the need to adequately protect the bitstreams against losses. The results also showed that the degradations in V-PCC occupancy and geometry sub-bitstreams have significantly higher (negative) impact on decoded point cloud subjective quality than degradations of the attribute sub-bitstream.

Learning-Based Near-Infrared Band Simulation with Applications on Large-Scale Landcover Classification.

Multispectral sensors are important instruments for Earth observation. In remote sensing applications, the near-infrared (NIR) band, together with the visible spectrum (RGB), provide abundant information about ground objects. However, the NIR band is typically not available on low-cost camera systems, which presents challenges for the vegetation extraction. To this end, this paper presents a conditional generative adversarial network (cGAN) method to simulate the NIR band from RGB bands of Sentinel-2 multispectral data. We adapt a robust loss function and a structural similarity index loss (SSIM) in addition to the GAN loss to improve the model performance. With 45,529 multi-seasonal test images across the globe, the simulated NIR band had a mean absolute error of 0.02378 and an SSIM of 89.98%. A rule-based landcover classification using the simulated normalized difference vegetation index (NDVI) achieved a Jaccard score of 89.50%. The evaluation metrics demonstrated the versatility of the learning-based paradigm in remote sensing applications. Our simulation approach is flexible and can be easily adapted to other spectral bands.

A robust motion correction technique for infrared thermography during awake craniotomy.

PURPOSE: Intraoperative infrared thermography is an emerging technique for image-guided neurosurgery, whereby physiological and pathological processes result in temperature changes over space and time. However, motion during data collection leads to downstream artifacts in thermography analyses. We develop a fast, robust technique for motion estimation and correction as a preprocessing step for brain surface thermography recordings. METHODS: A motion correction technique for thermography was developed which approximates the deformation field associated with motion as a grid of two-dimensional bilinear splines (Bispline registration), and a regularization function was designed to constrain motion to biomechanically feasible solutions. The performance of the proposed Bispline registration technique was compared to phase correlation, a band-stop filter, demons registration, and the Horn-Schunck and Lucas-Kanade optical flow techniques. RESULTS: All methods were analyzed using thermography data from ten patients undergoing awake craniotomy for brain tumor resection, and performance was compared using image quality metrics. The proposed method had the lowest mean-squared error and the highest peak-signal-to-noise ratio of all methods tested and performed slightly worse than phase correlation and Demons registration on the structural similarity index metric (p < 0.01, Wilcoxon signed-rank test). Band-stop filtering and the Lucas-Kanade method were not strong attenuators of motion, while the Horn-Schunck method was well-performing initially but weakened over time. CONCLUSION: Bispline registration had the most consistently strong performance out of all the techniques tested. It is relatively fast for a nonrigid motion correction technique, capable of processing ten frames per second, and could be a viable option for real-time use. Constraining the deformation cost function through regularization and interpolation appears sufficient for fast, monomodal motion correction of thermal data during awake craniotomy.

Properties of the SSIM metric in medical image assessment: correspondence between measurements and the spatial frequency spectrum.

In radiological imaging, the acquisition of the required diagnostic image quality under optimized conditions is important. Although techniques based on structural similarity (SSIM) have been investigated, concerns have been raised regarding their application to medical images. This study aims to clarify the properties of SSIM as an image quality index in medical images, focusing on digital radiography and verifying the correspondence between the evaluation results obtained by SSIM and the frequency spectrum. The analysis target was chest X-ray images of a human-body phantom. Various types of processing were applied to the images, and several regions of interest (ROIs) were used in local areas for analysis. The SSIM was measured using unprocessed data as a reference while changing the calculation parameters, and the spatial frequency spectrum of each local region was analyzed. Thus, a significant effect of ROI size was observed when calculating the SSIM. This indicates that larger the ROI size leads to SSIM values closer to 1 for all analysis conditions. In addition, a relationship between the size of the ROI in the analysis and the frequency components is demonstrated. It was shown that careful attention should be paid to the structures included in the ROI, and parameter settings should be reconsidered. Furthermore, when using SSIM to assess medical images, a multiscale SSIM method obtained by changing the ROI size would be useful.

The Impact of Excitation Periods on the Outcome of Lock-In Thermography.

Thermal imaging is a non-destructive test method that uses an external energy source, such as a halogen lamp or flash lamp, to excite the material under test and measure the resulting temperature distribution. One of the important parameters of lock-in thermography is the number of excitation periods, which is used to calculate a phase image that shows defects or inhomogeneities in the material. The results for multiple periods can be averaged, which leads to noise suppression, but the use of a larger number of periods may cause an increase in noise due to unsynchronization of the camera and the external excitation source or may lead to heating and subsequent damage to the sample. The phase image is the most common way of representing the results of lock-in thermography, but amplitude images and complex images can also be obtained. In this study, eight measurements were performed on different samples using a thermal pulse source (flash lamp and halogen lamp) with a period of 120 s. For each sample, five phase images were calculated using different number of periods, preferably one to five periods. The phase image calculated from one period was used as a reference. To determine the effect of the number of excitation periods on the phase image, the reference phase image for one period was compared with the phase images calculated using multiple periods using the structural similarity index (SSIM) and multi-scale SSIM (MS-SSIM).

Improving the brain image resolution of generalized q-sampling MRI revealed by a three-dimensional CNN-based method.

Background: Understanding neural connections facilitates the neuroscience and cognitive behavioral research. There are many nerve fiber intersections in the brain that need to be observed, and the size is between 30 and 50 nanometers. Improving image resolution has become an important issue for mapping the neural connections non-invasively. Generalized q-sampling imaging (GQI) was used to reveal the fiber geometry of straight and crossing. In this work, we attempted to achieve super-resolution with a deep learning method on diffusion weighted imaging (DWI). Materials and methods: A three-dimensional super-resolution convolutional neural network (3D SRCNN) was utilized to achieve super-resolution on DWI. Then, generalized fractional anisotropy (GFA), normalized quantitative anisotropy (NQA), and the isotropic value of the orientation distribution function (ISO) mapping were reconstructed using GQI with super-resolution DWI. We also reconstructed the orientation distribution function (ODF) of brain fibers using GQI. Results: With the proposed super-resolution method, the reconstructed DWI was closer to the target image than the interpolation method. The peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM) were also significantly improved. The diffusion index mapping reconstructed by GQI also had higher performance. The ventricles and white matter regions were much clearer. Conclusion: This super-resolution method can assist in postprocessing low-resolution images. With SRCNN, high-resolution images can be effectively and accurately generated. The method can clearly reconstruct the intersection structure in the brain connectome and has the potential to accurately describe the fiber geometry on a subvoxel scale.

Finding the Best Lighting Mode for Daguerreotype, Ambrotype, and Tintype Photographs and Their Deterioration on the Cruse Scanner Based on Selected Methods.

In the digital heritage field, the accurate reproduction of hard-to-photograph items, such as daguerreotypes, ambrotypes, and tintypes, is an ongoing challenge. Industrial contactless sensors offer the potential to improve the quality of scanned images, but their capabilities and limitations have not been fully explored. In our research, a dataset of 48 scans was created using the hi-tech industrial contactless sensor CRUSE. Moreover, 3 rare original photographs were scanned in 16 different modes, the most suitable images were determined by specialists in the restoration, and validated through experiments involving eye-tracking, multiple computer vision, and image processing methods. Our study identified the Cruse scanning modes, which can be utilized to achieve the most accurate digital representation of scanned originals. Secondly, we proposed several methods for highlighting the degradation and minor scratches on photographs that otherwise might not be detected by the restorer's naked eye. Our findings belong to four overlapping areas, i.e., image understanding, digital heritage, information visualization, and industrial sensors research. We publish the entire dataset under the CC BY-NC 4.0 license. The CRUSE sensor shows promise as a tool for improving the quality of scanned images of difficult-to-photograph items. Further research is necessary to fully understand its capabilities and limitations in this context.

A Lightweight Low-dose PET Image Super-resolution Reconstruction Method based on Convolutional Neural Network.

BACKGROUND: PET imaging is one of the most widely used neurological disease screening and diagnosis techniques. AIMS: Since PET involves the radiation and tolerance of different people, the improvement that has always been focused on is to cut down radiation, in the meantime, ensuring that the generated images with low-dose tracer and generated images with standard-dose tracer have the same details of images. METHODS: We propose a lightweight low-dose PET super-resolution network (SRPET-Net) based on a convolutional neural network. In this research, We propose a method for accurately recovering highresolution (HR) PET images from low-resolution (LR) PET images. The network learns the details and structure of the image between low-dose PET images and standard-dose PET images and, afterward, reconstructs the PET image by the trained network model. RESULTS: The experiments indicate that the SRPET-Net can achieve a superior peak signal-to-noise ratio (PSNR) and structural similarity index measurement (SSIM) values. Moreover, our method has less memory consumption and lower computational cost. CONCLUSION: In our follow-up work, the technology can be applied to medical imaging in many different directions.

OrbitNet-A fully automated orbit multi-organ segmentation model based on transformer in CT images.

The delineation of orbital organs is a vital step in orbital diseases diagnosis and preoperative planning. However, an accurate multi-organ segmentation is still a clinical problem which suffers from two limitations. First, the contrast of soft tissue is relatively low. It usually cannot clearly show the boundaries of organs. Second, the optic nerve and the rectus muscle are difficult to distinguish because they are spatially adjacent and have similar geometry. To address these challenges, we propose the OrbitNet model to automatically segment orbital organs in CT images. Specifically, we present a global feature extraction module based on the transformer architecture called FocusTrans encoder, which enhance the ability to extract boundary features. To make the network focus on the extraction of edge features in the optic nerve and rectus muscle, the SA block is used to replace the convolution block in the decoding stage. In addition, we use the structural similarity measure (SSIM) loss as a part of the hybrid loss function to learn the edge differences of the organs better. OrbitNet has been trained and tested on the CT dataset collected by the Eye Hospital of Wenzhou Medical University. The experimental results show that our proposed model achieved superior results. The average Dice Similarity Coefficient (DSC) is 83.9%, the value of average 95% Hausdorff Distance (HD95) is 1.62 mm, and the value of average Symmetric Surface Distance (ASSD) is 0.47 mm. Our model also has good performance on the MICCAI 2015 challenge dataset.

An image forensic technique based on JPEG ghosts.

The unprecedented growth in the easy availability of photo-editing tools has endangered the power of digital images. An image was supposed to be worth more than a thousand words, but now this can be said only if it can be authenticated or the integrity of the image can be proved to be intact. In this paper, we propose a digital image forensic technique for JPEG images. It can detect any forgery in the image if the forged portion called a ghost image is having a compression quality different from that of the cover image. It is based on resaving the JPEG image at different JPEG qualities, and the detection of the forged portion is maximum when it is saved at the same JPEG quality as the cover image. Also, we can precisely predict the JPEG quality of the cover image by analyzing the similarity using Structural Similarity Index Measure (SSIM) or the energy of the images. The first maxima in SSIM or the first minima in energy correspond to the cover image JPEG quality. We created a dataset for varying JPEG compression qualities of the ghost and the cover images and validated the scalability of the experimental results. We also, experimented with varied attack scenarios, e.g. high-quality ghost image embedded in low quality of cover image, low-quality ghost image embedded in high-quality of cover image, and ghost image and cover image both at the same quality. The proposed method is able to localize the tampered portions accurately even for forgeries as small as 10 × 10 sized pixel blocks. Our technique is also robust against other attack scenarios like copy-move forgery, inserting text into image, rescaling (zoom-out/zoom-in) ghost image and then pasting on cover image.

Early Glaucoma Detection by Using Style Transfer to Predict Retinal Nerve Fiber Layer Thickness Distribution on the Fundus Photograph.

Objective: We aimed to develop a deep learning (DL)-based algorithm for early glaucoma detection based on color fundus photographs that provides information on defects in the retinal nerve fiber layer (RNFL) and its thickness from the mapping and translating relations of spectral domain OCT (SD-OCT) thickness maps. Design: Developing and evaluating an artificial intelligence detection tool. Subjects: Pretraining paired data of color fundus photographs and SD-OCT images from 189 healthy participants and 371 patients with early glaucoma were used. Methods: The variational autoencoder (VAE) network training architecture was used for training, and the correlation between the fundus photographs and RNFL thickness distribution was determined through the deep neural network. The reference standard was defined as a vertical cup-to-disc ratio of ≥0.7, other typical changes in glaucomatous optic neuropathy, and RNFL defects. Convergence indicates that the VAE has learned a distribution that would enable us to produce corresponding synthetic OCT scans. Main Outcome Measures: Similarly to wide-field OCT scanning, the proposed model can extract the results of RNFL thickness analysis. The structural similarity index measure (SSIM) and peak signal-to-noise ratio (PSNR) were used to assess signal strength and the similarity in the structure of the color fundus images converted to an RNFL thickness distribution model. The differences between the model-generated images and original images were quantified. Results: We developed and validated a novel DL-based algorithm to extract thickness information from the color space of fundus images similarly to that from OCT images and to use this information to regenerate RNFL thickness distribution images. The generated thickness map was sufficient for clinical glaucoma detection, and the generated images were similar to ground truth (PSNR: 19.31 decibels; SSIM: 0.44). The inference results were similar to the OCT-generated original images in terms of the ability to predict RNFL thickness distribution. Conclusions: The proposed technique may aid clinicians in early glaucoma detection, especially when only color fundus photographs are available.

Denoising Single Images by Feature Ensemble Revisited.

Image denoising is still a challenging issue in many computer vision subdomains. Recent studies have shown that significant improvements are possible in a supervised setting. However, a few challenges, such as spatial fidelity and cartoon-like smoothing, remain unresolved or decisively overlooked. Our study proposes a simple yet efficient architecture for the denoising problem that addresses the aforementioned issues. The proposed architecture revisits the concept of modular concatenation instead of long and deeper cascaded connections, to recover a cleaner approximation of the given image. We find that different modules can capture versatile representations, and a concatenated representation creates a richer subspace for low-level image restoration. The proposed architecture's number of parameters remains smaller than in most of the previous networks and still achieves significant improvements over the current state-of-the-art networks.