PSNR Quality Research Articles

IMU-aided adaptive mesh-grid based video motion deblurring.

Motion blur is a problem that degrades the visual quality of images for human perception and also challenges computer vision tasks. While existing studies mostly focus on deblurring algorithms to remove uniform blur due to their computational efficiency, such approaches fail when faced with non-uniform blur. In this study, we propose a novel algorithm for motion deblurring that utilizes an adaptive mesh-grid approach to manage non-uniform motion blur with a focus on reducing the computational cost. The proposed method divides the image into a mesh-grid and estimates the blur point spread function (PSF) using an inertial sensor. For each video frame, the size of the grid cells is determined adaptively according to the in-frame spatial variance of blur magnitude which is a proposed metric for the blur non-uniformity in the video frame. The adaptive mesh-size takes smaller values for higher variances, increasing the spatial accuracy of the PSF estimation. Two versions of the adaptive mesh-size algorithm are studied, optimized for either best quality or balanced performance and computation cost. Also, a trade-off parameter is defined for changing the mesh-size according to application requirements. The experiments, using real-life motion data combined with simulated motion blur demonstrate that the proposed adaptive mesh-size algorithm can achieve 5% increase in PSNR quality gain together with a 19% decrease in computation time on the average when compared to the constant mesh-size method.

Comparison of different algorithms to improve the quality of Image

Super-resolution (SR) become the most important task in image processing because of requirement of high quality of images. It becomes popular in technology like computer vision and pattern recognition. In Single Image Super-resolution (SR) uses the sequences of single low resolution (LR) images from high resolution (HR) to improve the quality of image. The SR is an image processing problem which aims to generate a high resolution (HR) image from low resolution input image. The SR is difficult because of the missing information in the given LR image. Thus in SR it is very important to boost the reconstruction performance so that, there is need of effective prior knowledge. The prior knowledge of image will solve the problem to improve the quality of generated image. The main aim of SR is to generate the HR image with good visual perception as similar as original image. The paper compare two methods of SR that is bi-linear method, and direction group sparsity of image method that is proposed one. These methods will evaluate the accuracy, subjective visual effect, quality of image, time complexity and PSNR, etc. These calculations will enhance nature of picture like better visual impact quality, lower reconstruction error and adequate calculation proficiency, and so on

A new multi-picture architecture for learned video deinterlacing and demosaicing with parallel deformable convolution and self-attention blocks

Despite the fact real-world video deinterlacing and demosaicing are well-suited to supervised learning from synthetically degraded data because the degradation models are known and fixed, learned video deinterlacing and demosaicing have received much less attention compared to denoising and super-resolution tasks. We propose a new multi-picture architecture for video deinterlacing or demosaicing by aligning multiple supporting pictures with missing data to a reference picture to be reconstructed, benefiting from both local and global spatio-temporal correlations in the feature space using modified deformable convolution blocks and a novel residual efficient top-k self-attention (kSA) block, respectively. Separate reconstruction blocks are used to estimate different types of missing data. Our extensive experimental results, on synthetic or real-world datasets, demonstrate that the proposed novel architecture provides superior results that significantly exceed the state-of-the-art for both tasks in terms of PSNR, SSIM, and perceptual quality. Ablation studies are provided to justify and show the benefit of each novel modification made to the deformable convolution and residual efficient kSA blocks. Code is available: https://github.com/KUIS-AI-Tekalp-Research-Group/Video-Deinterlacing.

Quad Bayer Joint Demosaicing and Denoising Based on Dual Encoder Network with Joint Residual Learning

The recent imaging technology Quad Bayer CFA brings better imaging PSNR and higher visual quality compared to traditional Bayer CFA, but also serious challenges for demosaicing and denoising during the ISP pipeline. In this paper, we propose a novel dual encoder network, namely DRNet, to achieve joint demosaicing and denoising for Quad Bayer CFA. The dual encoders are carefully designed in that one is mainly constructed by a joint residual block to jointly estimate the residuals for demosaicing and denoising separately. In contrast, the other one is started with a pixel modulation block which is specially designed to match the characteristics of Quad Bayer pattern for better feature extraction. We demonstrate the effectiveness of each proposed component through detailed ablation investigations. The comparison results on public benchmarks illustrate that our DRNet achieves an apparent performance gain~(0.38dB to the 2nd best) from the state-of-the-art method and balances performance and efficiency well. The experiments on real-world images show that the proposed method could enhance the reconstruction quality from the native ISP algorithm.

Contrastive Learning for Low-Light Raw Denoising (Student Abstract)

Image/video denoising in low-light scenes is an extremely challenging problem due to limited photon count and high noise. In this paper, we propose a novel approach with contrastive learning to address this issue. Inspired by the success of contrastive learning used in some high-level computer vision tasks, we bring in this idea to the low-level denoising task. In order to achieve this goal, we introduce a new denoising contrastive regularization (DCR) to exploit the information of noisy images and clean images. In the feature space, DCR makes the denoised image closer to the clean image and far away from the noisy image. In addition, we build a new feature embedding network called Wnet, which is more effective to extract high-frequency information. We conduct the experiments on a real low-light dataset that captures still images taken on a moonless clear night in 0.6 millilux and videos under starlight (no moon present). The results show that our method can achieve a higher PSNR and better visual quality compared with existing methods.

Adaptively hybrid fractal image coding

AbstractAn adaptively hybrid method was proposed to improve the performance of fractal coding methods. First, it is found that the range blocks with large variances (RBLVs) play a crucial role in degrading decoded images, and the effect of the remaining range blocks with small variances (RBSVs) can be ignored. Then, an adaptive method was proposed to divide the range blocks into the above two categories: RBLVs and RBSVs. Second, RBLVs were designed to be encoded in an extended domain block pool (EDBP). Then, better block‐matching effect can be obtained, which will result in better decoded image quality. Further, the no‐search fractal encoding method is adopted for RBSVs to achieve faster encoding speed and fewer bits per pixel (bpp). Finally, four fractal coding methods were adopted to assess the performance of the proposed method. Experimental results show that compared with the previous methods, the PSNR quality of decoded images in the proposed method can be improved by about 0.15–0.4 dB, about 20%–35% of the total computations in encoding process can be saved, and about 0.2 bpp can be saved. Moreover, under the same decoding time, the proposed method can achieve comparable or smaller deviations regarding the decoded image.

Single image super-resolution with denoising diffusion GANS

Single image super-resolution (SISR) refers to the reconstruction from the corresponding low-resolution (LR) image input to a high-resolution (HR) image. However, since a single low-resolution image corresponds to multiple high-resolution images, this is an ill-posed problem. In recent years, generative model-based SISR methods have outperformed conventional SISR methods in performance. However, the SISR methods based on GAN, VAE, and Flow have the problems of unstable training, low sampling quality, and expensive computational cost. These models also struggle to achieve the trifecta of diverse, high-quality, and fast sampling. In particular, denoising diffusion probabilistic models have shown impressive variety and high quality of samples, but their expensive sampling cost prevents them from being well applied in the real world. In this paper, we investigate the fundamental reason for the slow sampling speed of the SISR method based on the diffusion model lies in the Gaussian assumption used in the previous diffusion model, which is only applicable for small step sizes. We propose a new Single Image Super-Resolution with Denoising Diffusion GANS (SRDDGAN) to achieve large-step denoising, sample diversity, and training stability. Our approach combines denoising diffusion models with GANs to generate images conditionally, using a multimodal conditional GAN to model each denoising step. SRDDGAN outperforms existing diffusion model-based methods regarding PSNR and perceptual quality metrics, while the added latent variable Z solution explores the diversity of likely HR spatial domain. Notably, the SRDDGAN model infers nearly 11 times faster than diffusion-based SR3, making it a more practical solution for real-world applications.

Diving into Clarity: Restoring Underwater Images using Deep Learning

In this paper we propose a learning-based restoration approach to learn the optimal parameters for enhancing the quality of different types of underwater images and apply a set of intensity transformation techniques to process raw underwater images. The methodology comprises two steps. Firstly, a Convolutional Neural Network (CNN) Regression model is employed to learn enhancing parameters for each underwater image type. Trained on a diverse dataset, the CNN captures complex relationships, enabling generalization to various underwater conditions. Secondly, we apply intensity transformation techniques to raw underwater images. These transformations collectively compensate for visual information loss due to underwater degradation, enhancing overall image quality. In order to evaluate the performance of our proposed approach, we conducted qualitative and quantitative experiments using well-known underwater image datasets (U45 and UIEB), and using the proposed challenging dataset composed by 276 underwater images from the Amazon region (AUID). The results demonstrate that our approach achieves an impressive accuracy rate in different underwater image datasets. For U45 and UIEB datasets, regarding PSNR and SSIM quality metrics, we achieved 26.967, 0.847, 27.299 and 0.793, respectively. Meanwhile, the best comparison techniques achieved 26.879, 0.831, 27.157 and 0.788, respectively.

Enhancing the Visual Effectiveness of Overexposed and Underexposed Images in Power Marketing Field Operations Using Gray Scale Logarithmic Transformation and Histogram Equalization

Abstract In this paper, we propose an adaptive gamma transform that adjusts the local values of bright and dark parts to enhance the effect of low-illumination images, thereby improving the light component. We then apply diff texture enhancement to enhance the contrast of images processed by the Retinex algorithm, thereby optimizing the perception of overexposed and underexposed imagery. Analyze the effect of image brightness enhancement based on a nonlinear transformation combined with the LOL dataset. Use PSNR and SSIM image quality evaluation criteria to analyze the visual effect of improving low-illumination images based on Retinex theory. Create a dataset of power marketing field operation inspection images and examine the effects of overexposure and underexposure image processing on four types of images: high-voltage towers, transmission lines, high-voltage fixtures, and high-voltage wireframes, using the low-light image texture fusion algorithm based on Retinex theory. Overall, this paper’s algorithm and the three DeblurGAN and DMCNN models achieve the effect of deblurring overexposed and underexposed power marketing field operation inspection images. From the local details, the model in this paper has a better effect on the de-exposure of the image, which can provide effective help for the electric power staff to understand the situation of the electric power marketing operation site and has strong practicality.

A Simplified Convex Optimization Model for Image Restoration with Multiplicative Noise.

In this paper, we propose a novel convex variational model for image restoration with multiplicative noise. To preserve the edges in the restored image, our model incorporates a total variation regularizer. Additionally, we impose an equality constraint on the data fidelity term, which simplifies the model selection process and promotes sparsity in the solution. We adopt the alternating direction method of multipliers (ADMM) method to solve the model efficiently. To validate the effectiveness of our model, we conduct numerical experiments on both real and synthetic noise images, and compare its performance with existing methods. The experimental results demonstrate the superiority of our model in terms of PSNR and visual quality.

MR image reconstruction using iterative up and downsampling network

Magnetic resonance imaging (MRI) is one of the most significant modalities in medical imaging. It suffers from rather lengthy acquisition times, which lead to prolonged patient restriction, patient discomfort, and imaging artifacts. Hence, satisfactory MRI reconstruction from undersampled data sequences constitutes an important research problem. With the advances in deep learning (DL), a plethora of new models have been proposed to solve the MRI reconstruction problem using deep networks. On the other hand, single image super-resolution (SR) is another well-studied field that benefited from the success of DL, and it has applications in various imaging modalities. SR is the process of recovering a high-resolution image from a low-resolution image. SR models work on low-resolution images to recover missing details. MR image reconstruction on the other hand is a battle to eliminate the aliasing artifacts which originate from data downsampling. The motivation for the proposed work is based on the premise that SR approaches can possibly get adapted to MR image reconstruction. Hence in this study, inspired by the great success of deep SR networks, we customize an architecture introduced in SR setting to MRI reconstruction. This novel approach uses the iterative up and downsampling framework labeled as Iterative Up and Down Network (IUDN) for MRI reconstruction. We design two variants of the proposed network with different number of scale factors. We present extensive simulations for the proposed architecture using multiple k-space undersampling ratios. The simulation results indicate cutting edge MRI reconstruction performance for the proposed models. The networks were trained on fastMRI dataset and tested on both fastMRI and IXI datasets to show the robustness of the method. The proposed models achieved improved results in terms of PSNR and visual quality of the reconstructed image when compared to some recent state-of-the-art solutions for MRI reconstruction.

End-to-End Depth-Guided Relighting Using Lightweight Deep Learning-Based Method.

Image relighting, which involves modifying the lighting conditions while preserving the visual content, is fundamental to computer vision. This study introduced a bi-modal lightweight deep learning model for depth-guided relighting. The model utilizes the Res2Net Squeezed block's ability to capture long-range dependencies and to enhance feature representation for both the input image and its corresponding depth map. The proposed model adopts an encoder-decoder structure with Res2Net Squeezed blocks integrated at each stage of encoding and decoding. The model was trained and evaluated on the VIDIT dataset, which consists of 300 triplets of images. Each triplet contains the input image, its corresponding depth map, and the relit image under diverse lighting conditions, such as different illuminant angles and color temperatures. The enhanced feature representation and improved information flow within the Res2Net Squeezed blocks enable the model to handle complex lighting variations and generate realistic relit images. The experimental results demonstrated the proposed approach's effectiveness in relighting accuracy, measured by metrics such as the PSNR, SSIM, and visual quality.

Predictive and Adaptive Deep Coding for Wireless Image Transmission in Semantic Communication

Semantic communication is a newly emerged communication paradigm that exploits deep learning (DL) models to realize communication processes like source coding and channel coding. Recent advances have demonstrated that DL-based joint source-channel coding (DeepJSCC) can achieve exciting data compression and noise-resiliency performances for wireless image transmission tasks, especially in environments with low channel signal-to-noises (SNRs). However, existing DeepJSCC-based semantic communication frameworks still cannot achieve adaptive code rates for different channel SNRs and image contents, which reduces its flexibility and bandwidth efficiency. In this paper, we propose a predictive and adaptive deep coding (PADC) framework for realizing flexible code rate optimization with a given target transmission quality requirement. PADC is realized by a variable code length enabled DeepJSCC (DeepJSCC-V) model for realizing flexible code length adjustment, an Oracle Network (OraNet) model for predicting peak-signal-to-noise (PSNR) value for an image transmission task according to its contents, channel signal to noise ratio (SNR) and the compression ratio (CR) value, and a CR optimizer aims at finding the minimal data-level or instance-level CR with a PSNR quality constraint. By using the above three modules, PADC can transmit the image data with minimal CR, which greatly increases bandwidth efficiency. Simulation results demonstrate that the proposed DeepJSCC-V model can achieve similar PSNR performances compared with the state-of-the-art Attention-based DeepJSCC (ADJSCC) model, and the proposed OraNet model is able to predict high-quality PSNR values with an average error lower than 0.5dB. Results also demonstrate that the proposed PADC can use nearly minimal bandwidth consumption for wireless image transmission tasks with different channel SNR and image contents, at the same time guaranteeing the PSNR constraint for each image data.

Super-Resolution Neural Network Models Comparison Metod for Brain MRI Images Based on PSNR and SSIM Metrics

The diagnosis of many diseases is largely possible thanks to MRI(Magnetic Resonance Imaging). This technology allows to study internal organs of the patient: the brain, spine, bones, joints, vessels and etc. The resolution of the MRI image is limited due to various factors: movement of the patient during the scan, the continuous movement of internal organs. The higher the quality of the MRI image, the longer it takes to scan. For more accurate diagnostics it is possible to increase resolution of the yielded images. This is achieved by using SISR(Single Image Super Resolution) algorithms, which allow you to obtain images with increased resolution from a single input image. In this paper the idea of the image super-resolution algorithms is presented, various forms of the problem and solutions to it are provided. The advantages of the SISR algorithms are described. The relevance of this task in the field of medical MRI images is explained. Metrics for comparing image quality PSNR and SSIM are given and described. A dataset for testing is presented. The stage of data preparation is described: the principle of selecting images from a set of datasets, converting data into the required format, compressing images to obtain input data for selected neural network models. The PSNR, SSIM metrics of two neural network models mDCSRN and FAWDN are measured on equally prepared input data. The comparison results are presented in the form of images and averaged data for the entire sample is stored in the table.

A Novel L-CLAHE-Based Intensification Filter for Enhancement of Underwater Images and Pipeline Tracking

The high contrast degraded underwater images often demand extensive pre-processing to attain reliable information for object detection and pipeline tracking. The prototype system developed employs CLAHE, a variant of adaptive histogram equalization, with an adaptive clip limit based on entropy. To yield an enhanced image, CLAHE is applied to the L* channel alone, resulting in improved L- CLAHE and better results. Validation of the improved L-CLAHE method is carried out using real-life underwater images captured using DTG2 Pro ROV from Kochi backwaters based on PSNR, SSIM, EPI, and PFOM quality measures and outperforms the conventional histogram equalization and adaptive histogram equalization techniques. These images are subjected to intensification and filtering for further enhancement as well as object detection and pipeline tracking applications. The proposed algorithm is compared with CLAHE, MSRCR, and L-CIF methods. The improved L-CIF algorithm has a high entropy value compared to the existing algorithms while the EME metric has been evaluated close to unity indicating the improved performance of the fast-guided filter. The method developed also helps in improving the average pixel intensity of the input images.

QUALITY OF VIDEO RENDERING TECHNIQUES USING ARTIFICIAL INTELLIGENCE

In this paper, we propose a novel method that makes use of artificial intelligence to determine in a quick and accurate manner which bitrate ladder is best suited to each specific video scenario. This method is included as part of our overall contribution to this body of research. To accomplish fast entropy-based scene recognition using the artificial intelligence technique, a CNN model is utilised as part of the overall strategy. We were able to significantly reduce the amount of processing time necessary to recognise the scenes because we were dealing with versions of the video sequences that had both a lower quality and a lower bitrate. This allowed us to work more quickly. We first generated a training dataset that was large enough to train a convolutional neural network utilising the x264 video codec, and then we used that dataset to generate multiple encodings with varying bitrates, presets, and resolutions. The training dataset was created using the x264 video codec. As a result of the research that we carried out, we concluded that a particular collection of input features for the CNN model can be used to acquire a more accurate prediction of the level of video quality that will be produced. By predicting the PSNR quality measure for the segments, the suggested CNN model brings down the MAE and MSE to 0.2 and 0.05, respectively. This is accomplished by reducing the number of segments. This serves to reduce the amount of error overall.

High-frequency Normalizing Flow for Image Rescaling.

It is desirable to develop efficient image rescaling methods to transmit digital images with different resolutions between devices and assure visual quality. In image downscaling, the inevitable loss of high-frequency information makes the reverse upscaling highly ill-posed. Recent approaches focus on joint learning of image downscaling and upscaling (e.g., rescaling). However, existing methods still fail to recover satisfactory high-frequency signals when upscaling. To solve it, we propose high-frequency flow (HfFlow), which learns the distribution of high-frequency signals during rescaling. HfFlow is an overall invertible framework with a conditional flow on the high-frequency space to compensate for the information lost during downscaling. To facilitate finding the optimal upscaling solution, we introduce a reference low-resolution (LR) manifold and propose a cross-entropy Gaussian loss (CGloss) to force the downscaled manifold closer to the reference LR manifold and simultaneously fulfill recovering missing details. HfFlow can be generalized to other scale transformation tasks such as image colorization with its excellent rescaling capacity. Qualitative and quantitative experimental evaluations demonstrate that HfFlow restores rich high-frequency details and outperforms state-of-the-art rescaling methods in PSNR, SSIM, and perceptual quality metrics.

Nonconvex model for mixing noise with fractional-order regularization

<p style='text-indent:20px;'>In this paper, we propose a restoration model for image degraded by different kinds of blur and mixing Gaussian-impulse noise. Our model consist of a nonconvex Exponential-Type (ET) function, a <inline-formula><tex-math id="M1">\begin{document}$ L_{2} $\end{document}</tex-math></inline-formula>-norm data-fitting term and a fractional-order Besov norm as the regularization term. We employ the proximal linearized minimization (PLM) algorithm and alternating direction method of multipliers (ADMM) algorithm to solve our proposed minimization model, convergence analysis is carried out. The experimental outcomes demonstrate that the restored images by our proposed method are better than those existing relative methods in terms of PSNR, SSIM values and visual quality.</p>

Protecting the Distribution of Color Images via Inverse Colorization, Visible-Imperceptible Watermarking and Reversible Data Hiding

Invertible color-to-gray algorithms are good alternatives for protecting color images in storage, transmission, or limited-access environments. These approaches hide color information in a gray-scale version of an image. Thus, illegal recovery of the original color becomes a challenging task, and only authorized users can restore the colorized image. However, although the color is protected, the original structure of the image remains vulnerable to illegal usage. In this paper, we present a reversible distortion of the protected gray-scale image to make illegal reconstruction more difficult. In addition, we preserve the general visibility of the original content. Furthermore, a visible imperceptible watermark is embedded to protect ownership of the colorized image in public access. The watermark remains imperceptible in the colorized image and reveals the logo of the owner when the image is protected. We propose different levels of distortions obtaining mean PSNR qualities between 14.74 dB and 32.92 dB with respect to a reference gray-scale image. Furthermore, the mean PSNR quality of the colorized images remains between 38.87 dB and 41.55 dB.

A Novel Unbiased Deep Learning Approach (DL-Net) in Feature Space for Converting Gray to Color Image

Gray to Color conversion causes difficulties because of the nature of its intrinsic multi-modality. Despite recent significant advancements in this domain by numerous learning-based approaches, there still have two drawbacks: i) implausible color assignment and ii) contextual ambiguity. Recently deep learning models are being used for colorization as they outperform others. In a training image, desaturated color components are greater than saturated color components due to the larger background areas (clouds, pavement, dirt, walls, etc.) compared to the focused objects. This imbalanced feature representation biases the learning model in favor of major features. However, small regions with specific colors are the region of interest. To solve this problem, we proposed the Deep Localization Network (DL-Net) by modifying the mean squared error backpropagation algorithm. We compute chromatic component-based Local Losses (LLs) which are the primary component of the proposed DL-Net. The LL employs priority on rare semantic components of the original image features. It works to improve diverse-range dependency modeling in an effort to reduce contextual ambiguity and color leakage that promotes the production of more plausible coloring. With a number of current methodologies, we contrast our proposed approach. The experimental findings demonstrate that our proposed method produces good colorization of images and outperforms other methods in terms of SSIM, MSE, and PSNR quality criteria.