Loss Of Detail Research Articles

Low-Light Image Enhancement Algorithm Based on Deep Learning and Retinex Theory

To address the challenges of low-light images, such as low brightness, poor contrast, and high noise, a network model based on deep learning and Retinex theory is proposed. The model consists of three modules: image decomposition, illumination enhancement, and color restoration. In the image decomposition module, dilated convolutions and residual connections are employed to mitigate the issue of detail loss during the decomposition process. The illumination enhancement module utilizes a set of mapping curves to enhance the illumination map. The color restoration module employs a weighted fusion of a 3D lookup table (3DLUT) to mitigate color distortion in the images. The experimental results demonstrate that the proposed algorithm effectively improves the brightness and contrast of low-light images while addressing the issues of detail loss and color distortion. Compared to other algorithms, it achieves better subjective and objective evaluations.

A lightweight object detection framework for underwater imagery with joint image restoration and color transformation

Traditional underwater detectors struggle to match the demands of visual observation and icon detection because of the underwater optical pictures' color distortion, uneven illumination, and excessive noise levels. In this paper, a lightweight deep-water object detection network is proposed, which consists of three key components, namely image restoration, color transformation, and object detection. First, we propose a lightweight attention module for the problems of color distortion and uneven illumination in underwater images. This module pays more attention to the areas with a greater influence of underwater color when dealing with different types of poor-quality images. Second, to reduce the computational complexity of object detection, we propose a color conversion module based on the laplacian pyramid decomposition. This module incorporates multi-scale features to effectively address the loss of details in the brightest or darkest regions. Finally, we propose a multi-scale dense residual object detection network. The network applies a multi-scale kernel to an improved residual module to realize the extraction and fusion of feature information of multiple receptive fields. Furthermore, to deal with the problem of underwater small-sample object detection, we propose a mutually exclusive loss function. The mutual exclusion loss promotes the update of network parameters so that the classes with different characteristics of samples generated by the network are mutually exclusive. We ran tests on the Pycharm platform to show that the suggested lightweight joint underwater object identification algorithm outperformed the most sophisticated ones. The proposed system was put to the test in an intricate underwater setting, achieving 92.56 percent accuracy and 19.2 frames per second.

A Sea Fog Image Defogging Method Based on the Improved Convex Optimization Model

Due to the high fog concentration in sea fog images, serious loss of image details is an existing problem, which reduces the reliability of aerial visual-based sensing platforms such as unmanned aerial vehicles. Moreover, the reflection of water surface and spray can easily lead to overexposure of images, and the assumed prior conditions contained in the traditional fog removal method are not completely valid, which affects the restoration effectiveness. In this paper, we propose a sea fog removal method based on the improved convex optimization model, and realize the restoration of images by using fewer prior conditions than that in traditional methods. Compared with dark channel methods, the solution of atmospheric light estimation is simplified, and the value channel in hue–saturation–value space is used for fusion atmospheric light map estimation. We construct the atmospheric scattering model as an improved convex optimization model so that the relationship between the transmittance and a clear image is deduced without any prior conditions. In addition, an improved split-Bregman iterative method is designed to obtain the transmittance and a clear image. Our experiments demonstrate that the proposed method can effectively defog sea fog images. Compared with similar methods in the literature, our proposed method can actively extract image details more effectively, enrich image color and restore image maritime targets more clearly. At the same time, objective metric indicators such as information entropy, average gradient, and the fog-aware density evaluator are significantly improved.

Multi-Scale Feature Learning Convolutional Neural Network for Image Denoising.

Affected by the hardware conditions and environment of imaging, images generally have serious noise. The presence of noise diminishes the image quality and compromises its effectiveness in real-world applications. Therefore, in real-world applications, reducing image noise and improving image quality are essential. Although current denoising algorithms can somewhat reduce noise, the process of noise removal may result in the loss of intricate details and adversely impact the overall image quality. Hence, to enhance the effectiveness of image denoising while preserving the intricate details of the image, this article presents a multi-scale feature learning convolutional neural network denoising algorithm (MSFLNet), which consists of three feature learning (FL) modules, a reconstruction generation module (RG), and a residual connection. The three FL modules help the algorithm learn the feature information of the image and improve the efficiency of denoising. The residual connection moves the shallow information that the model has learned to the deep layer, and RG helps the algorithm in image reconstruction and creation. Finally, our research indicates that our denoising method is effective.

A Novel Deep Learning Model for Medical Image Segmentation with Convolutional Neural Network and Transformer.

Accurate segmentation of medical images is essential for clinical decision-making, and deep learning techniques have shown remarkable results in this area. However, existing segmentation models that combine transformer and convolutional neural networks often use skip connections in U-shaped networks, which may limit their ability to capture contextual information in medical images. To address this limitation, we propose a coordinated mobile and residual transformer UNet (MRC-TransUNet) that combines the strengths of transformer and UNet architectures. Our approach uses a lightweight MR-ViT to address the semantic gap and a reciprocal attention module to compensate for the potential loss of details. To better explore long-range contextual information, we use skip connections only in the first layer and add MR-ViT and RPA modules in the subsequent downsampling layers. In our study, we evaluated the effectiveness of our proposed method on three different medical image segmentation datasets, namely, breast, brain, and lung. Our proposed method outperformed state-of-the-art methods in terms of various evaluation metrics, including the Dice coefficient and Hausdorff distance. These results demonstrate that our proposed method can significantly improve the accuracy of medical image segmentation and has the potential for clinical applications. Illustration of the proposed MRC-TransUNet. For the input medical images, we first subject them to an intrinsic downsampling operation and then replace the original jump connection structure using MR-ViT. The output feature representations at different scales are fused by the RPA module. Finally, an upsampling operation is performed to fuse the features to restore them to the same resolution as the input image.

C-SwinIR: Dental CT super-resolution reconstruction fused with SwinIR

Super-resolution reconstruction (SR) of dental computed tomography (Dental CT) images is a innovative and challenging task. To address the limitations of Dental CT in obtaining high-resolution (HR) images due to equipment constraints and noise interference, we propose a Dental CT SR method called C-SwinIR based on SwinIR. Firstly, the self-calibrated convolutions network (SCNet) is introduced to solve the problem of detail loss in shallow feature graphs and improve the ability to recover details. Subsequently, the cross-shaped windows (CSWin) self-attention transformer structure is used to replace the original transformer structure, which improves the ability of the model to obtain context information. Eventually, the integration of efficient channel attention (ECA-Net) module effectively realizes the local cross-channel interaction, accelerates the model convergence and solves the problem of gradient explosion in training. Experimental results show that our proposed method is superior to the original SwinIR network by achieving a higher peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) by 0.287 and 0.003 respectively. Additionally, it decreases mean squared error (MSE) by 1.108. By using this method, clinicians can obtain clearer details and textures from Dental CT, which effectively assist in diagnosis.

Enhancing Image Clarity: A Non-Local Self-Similarity Prior Approach for a Robust Dehazing Algorithm

When light propagates in foggy weather, it is affected and scattered by suspended particles in the air. As a result, images taken in this environment often suffer from blurring, reduced contrast, loss of details, and other issues. The primary challenge in dehazing images is to estimate the transmission coefficient map in the atmospheric degradation model. In this paper, we propose a dehazing algorithm based on the optimization of the “haze-line” prior and non-local self-similarity prior. First, we divided the input haze image into small blocks and used the nearest neighbor classification algorithm to cluster the small patches, which were referred to as “patch-lines”. Based on the characteristics of these “patch-lines”, we could estimate the transmission coefficient map for the image. We then applied the transmission map to a weighted least squares filter to smooth it. Finally, we calculated the clear image using the haze degradation model. The experimental results demonstrate that our algorithm enhanced the image contrast and preserved the fine details, both qualitatively and quantitatively.

Destriping of Remote Sensing Images by an Optimized Variational Model.

Satellite sensors often capture remote sensing images that contain various types of stripe noise. The presence of these stripes significantly reduces the quality of the remote images and severely affects their subsequent applications in other fields. Despite the existence of many stripe noise removal methods in the research, they often result in the loss of fine details during the destriping process, and some methods even generate artifacts. In this paper, we proposed a new unidirectional variational model to remove horizontal stripe noise. The proposed model fully considered the directional characteristics and structural sparsity of the stripe noise, as well as the prior features of the underlying image, to design different sparse constraints, and the ℓp quasinorm was introduced in these constraints to better describe these sparse characteristics, thus achieving a more excellent destriping effect. Moreover, we employed the fast alternating direction method of multipliers (ADMM) to solve the proposed non-convex model. This significantly improved the efficiency and robustness of the proposed method. The qualitative and quantitative results from simulated and real data experiments confirm that our method outperforms existing destriping approaches in terms of stripe noise removal and preservation of image details.

InvFlow: Involution and multi-scale interaction for unsupervised learning of optical flow

The convolution neural network is still the main tool for extracting the image features and the motion features for most of the optical flow models. The convolution neural networks cannot model the long-range dependencies, and more details are lost in deeper layers. All the deficiencies in the extracted features affect the estimated flow. Therefore, in this work, we concentrated on optimizing the convolution neural network in both the encoder and decoder parts to improve the image and motion features. To enhance the image features, we utilize the involution to provide rich features and model the long-range dependencies. In addition, we propose a Multi-Scale-Interaction module which utilizes the self-attention to make an interaction between the feature scales to avoid detail loss. Additionally, we propose a Motion-Features-Optimization block that utilizes the deformable convolution to enhance the motion features. Our model achieves the state-of-the-art performance on Sintel and KITTI 2015 benchmarks.

Multi and hyperspectral image unmixing with spatial coherence by extended blind end-member and abundance extraction

Blind linear unmixing (BLU) methods decompose multi and hyperspectral datasets into end-members and abundance maps with an unsupervised perspective. However, due to measurement noise and model uncertainty, the estimated abundance maps could exhibit granularity, which causes a loss of detail that could be crucial in certain applications. To address this problem, in this paper, we present a BLU proposal that considers spatial coherence (SC) in the abundance estimates. The proposed BLU formulation is based on the extended blind end-member and abundance extraction (EBEAE) methodology, and is denoted as EBEAE-SC. In this proposed method, the energy functional of EBEAE-SC includes new variables, which are denoted as internal abundances, to induce SC in the BLU approach. The new formulation of the optimization problem is solved by a coordinate descent algorithm, constrained quadratic optimization, and the split Bregman formulation. We present a comprehensive validation process that considers synthetic and experimental datasets at different noise types and levels, and a comparison with five state-of-the-art BLU methods. In our results, EBEAE-SC can significantly decrease the granularity in the estimated abundances, without losing detail of the structures present in the multi and hyperspectral images. In addition, the resulting complexity of EBEAE-SC is analyzed and compared it to the original formulation of EBEAE, and also the numerical convergence of the resulting iterative process is evaluated. Hence, by our analysis, EBEAE-SC allows blind estimates of end-members and abundances in the studied datasets of diverse applications, producing linearly independent and non-negative end-members, as well as non-negative abundances, with lower estimation errors and computational times compared to five methodologies in the state-of-the-art.

SHDM-NET: Heat map detail guidance with image matting for industrial weld semantic segmentation network

Welding is widely used in metal components. The firm of weld components is very important in different applications, such as buildings, bridges, cars and airplanes, etc. Weld seam quality inspection is essential to ensure product quality. The area and shape of the weld seam are the basis for quality assessment. So the segmentation of the weld area is very important for quality assessment. To address the problem of segmentation of the weld seam region, a weld seam segmentation network based on heat map detail guidance with Matting is proposed in this paper, which provides a new idea for fine-grained segmentation of the weld seam region. The existing DCNN-based semantic segmentation algorithm model has a poor segmentation effect at the boundary and jagged segmentation boundary, which are unacceptable for the weld segmentation problem that requires clear and precise boundary positioning. To solve this problem, three innovations are made in this paper on the DCNN-based semantic segmentation network. (1) A heat map detail guidance module makes the segmentation boundary information focus on shallow features and enhances the representation of boundary information. (2) A segmentation head improvement method for fine-grained semantic segmentation is proposed. (3) In response to the loss of process details in the coding and decoding of the semantic segmentation network, resulting in poor segmentation boundary accuracy, this paper introduces a matting algorithm to calibrate the boundary of the weld seam segmentation region. Through many experiments on industrial weld data sets, the effectiveness of our method is demonstrated, and the MIoU (Mean Intersection over Union) reaches 96.32%. It is worth noting that this performance is comparable to human manual segmentation ( MIoU 96.38%).

Low-light image enhancement based onRetinex-Net with color restoration.

Low-light images often suffer from a variety of degradation problems such as loss of detail, color distortions, and prominent noise. In this paper, the Retinex-Net model and loss function with color restoration are proposed to reduce color distortion in low-light image enhancement. The model trains the decom-net and color recovery-net to achieve decomposition of low-light images and color restoration of reflected images, respectively. First, a convolutional neural network and the designed loss functions are used in the decom-net to decompose the low-light image pair into an optimal reflection image and illumination image as the input of the network, and the reflection image after normal light decomposition is taken as the label. Then, an end-to-end color recovery network with a simplified model and time complexity is learned and combined with the color recovery loss function to obtain the correction reflection map with higher perception quality, and gamma correction is applied to the decomposed illumination image. Finally, the corrected reflection image and the illumination image are synthesized to get the enhanced image. The experimental results show that the proposed network model has lower brightness-order-error (LOE) and natural image quality evaluator (NIQE) values, and the average LOE and NIQE values of the low-light dataset images can be reduced to 942 and 6.42, respectively, which significantly improves image quality compared with other low-light enhancement methods. Generally, our proposed method can effectively improve image illuminance and restore color information in the end-to-end learning process of low-light images.

LPS-Net: Lightweight Parallel Strategy Network for Underwater Image Enhancement

Underwater images are frequently subject to color distortion and loss of details. However, previous enhancement methods did not tackle these mixed degradations by dividing them into sub-problems that could be effectively addressed. Moreover, the parameters and computations required for these methods are usually costly for underwater equipment, which has limited power supply, processing capabilities, and memory capacity. To address these challenges, this work proposes a Lightweight Parallel Strategy Network (LPS-Net). Firstly, a Dual-Attention Enhancement Block and a Mirror Large Receptiveness Block are introduced to, respectively, enhance the color and restore details in degraded images. Secondly, we employed these blocks on parallel branches at each stage of LPS-Net, with the goal of achieving effective image color and detail rendering simultaneously. Thirdly, a Gated Fusion Unit is proposed to merge features from different branches at each stage. Finally, the network utilizes four stages of parallel enhancement, achieving a balanced trade-off between performance and parameters. Extensive experiments demonstrated that LPS-Net achieves optimal color enhancement and superior detail restoration in terms of visual quality. Furthermore, it attains state-of-the-art underwater image enhancement performance on the evaluation metrics, while using only 80.12 k parameters.

Hybrid feature fusion neural network integrating transformer for DCE-MRI super resolution

Dynamic Contrast-Enhanced Magnetic Resonance Imaging (DCE-MRI) is an essential medical imaging technique that provides multiple time-series images of the same anatomical structures and function without causing ionizing radiation. However, due to the physical nature of the DCE-MRI process, the scan time can be as long as tens of minutes, which severely affects the patient experience. Therefore, reducing scanning time has become a hot research topic, which super-resolution reconstruction from post-processing of the resulting low-resolution images to obtain high-resolution images. However, some existing methods exhibit general limitations in capturing long-range dependencies and loss of localized details. In this paper, we propose a Hybrid Feature Fusion Neural Network Integrating Transformer for DCE-MRI Super Resolution Reconstruction, we call it HybridF. HybridF combines Transformer and CNN, where both global dependency and low-level spatial details can be efficiently captured. Moreover, where a Hybrid Fusion Block (HFB) is proposed to effectively aggregate multi-level features from Transformer and CNN. The method is evaluated by the data collection of clinical DCE-MRI and public IXI dataset. The DCE-MRI dataset consists of 212 cases of patients images. The reconstruction quality of our network is higher than that of the SRCNN, EDSR and RCAN method (＋2.24 dB, ＋2.79, ＋1.09). It is demonstrated that our network outperforms state-of-the art approaches it will contribute to the development of DCE-MRI super-resolution reconstruction.

Remote sensing building damage assessment with a multihead neighbourhood attention transformer

ABSTRACT Most existing remote sensing disaster assessment methods rely on convolutional neural networks (CNNs). Although CNNs can extract effective semantic features, determining global spatial relationships remains limited due to the locality of convolutional operations. The recently developed transformer-based method can extract global information from images effectively by encoding image tokens. However, its consumption of computational resources varies markedly with increasing image resolution. In this paper, we propose a novel transformer-based neural network for disaster assessment problems. The network uses a multihead neighbourhood attention (MNA) transformer as the base layer of the encoder to achieve more efficient self-attention computation. In addition, the bitemporal feature fusion module (BFFM) performs differential enhancement and injects the change information to the decoder via skip connections. The multiscale tokenizer generates multiscale image tokens to mitigate the loss of detail during encoding. Experimental results on three datasets show that the proposed method outperforms existing methods.

MSPIF: Multi-stage progressive visible and infrared image fusion with structures preservation

Infrared and visible image fusion is one of the hottest research in computer vision to improve image quality. Traditional fusion methods suffer from detail loss, low resolution, and application scenarios limitation. To address these problems, this paper proposes a multi-stage progressive visible and infrared image fusion strategy (MSPIF), including the following key stages: Firstly, the visible image is enhanced using a weighted fusion visible image enhancement algorithm. Secondly, the infrared and enhanced visible images are both decomposed by a pre-trained network, namely Retinex_Net, to obtain their respective illumination and reflectance components. Thirdly, the reflectance components are decomposed by discrete wavelet transform (DWT). The low-frequency components are fused by a weighted information entropy fusion strategy while the high-frequency components are fused by a local energy fusion strategy. Such two fused parts contribute to the final fused reflectance component with inverse DWT (IDWT). Meanwhile, the illumination components are fused using a multi-scale fusion strategy based on the Laplace pyramid. Finally, the improved result is generated by the fused reflectance and illumination components based on Retinex theory. Quantitative and qualitative results of experiments on the TNO, KAIST, and ROAD datasets show that the proposed MSPIF is effective and achieves good results with structures preservation, even is superior to existing methods.

Restoration and Enhancement of Fuzzy Defect Image Based on Neural Network

<p>In contrast enhancement of fuzzy defect image, details loss and noise expansion are east to occur, which brings difficulties to subsequent image analysis and defect recognition. Therefore, a fuzzy defect image restoration and enhancement method based on neural network is proposed. A double fusion neural network composed of a depth generation network and a discrimination network is designed. The residual of the denoised fuzzy image and the real image is output by the network, which is input into the discrimination network together with the real image, and the difference between the two is judged by the total loss function. To solve the problem of pixel coordinate value of fuzzy defect image, neural network is used to build a fast correction algorithm. Therefore, a fuzzy image restoration and enhancement method based on neural network is proposed to improve the image quality. By reconstructing the resolution of fuzzy defect image, a hierarchical enhancement method of fuzzy defect image region is constructed to achieve fuzzy defect image restoration and enhancement. The results show that the proposed method has high image processing ability in restoration and enhancement of fuzzy defect images. The fitting value of neural network is 0.92, which is significantly higher than that of the other two methods, indicating that the image restoration and enhancement method based on neural network has higher accuracy. Therefore, the restoration and enhancement method of fuzzy defect image based on neural network has a good restoration and enhancement effect, and can effectively meet the actual needs of people for high-quality images.</p> <p> </p>

Hyperspectral image destriping and denoising from a task decomposition view

The generalized mathematical model for HSI denoising or destriping lacks stability and uniqueness properties, failing to accurately portray the distribution and effects of stripes. Solutions following such a model would inevitably result in excessive destriping of strip-free areas, leading to the loss of texture detail. To remedy the above deficiencies, we reformulate the destriping task and introduce a novel solution from the task decomposition view. It is broken down into auxiliary sub-tasks involving stripe mask detection, stripe intensity estimation, and HSI restoration, which greatly reduces the difficulty of solving such an ill-posed problem. Based on this, we adopt a sequential multi-task learning framework and propose a stripes location-dependent restoration network, termed SLDR, which integrates the distribution and intensity features of stripes to achieve accurate destriping and high-fidelity restoration. Furthermore, we design a stripe attribute-aware estimator and a weighted total variation loss function to capture the unique properties of stripes and adaptively adjust the restoration weights of striped and non-striped regions. Extensive evaluation and comprehensive ablation studies on synthetic and practical scenes show the effectiveness and superiority of our model and architecture.

Terahertz 3D point cloud imaging for complex targets.

The reconstruction of complex targets using terahertz technology is often hindered by diffraction and interference of electromagnetic waves, leading to the loss of fine target details. In this research article, we have introduced a terahertz synthetic aperture radar (SAR) imaging method that integrates an iterative closest point (ICP) algorithm, referred to as SAR-ICP, to achieve accurate reconstruction of intricate target structures. To accomplish this, multiple sets of point cloud data are acquired by varying the illumination viewpoint. The ICP algorithm is then employed to align and fuse these datasets, resulting in the generation of high-quality three-dimensional (3D) images. The experimental results validate the effectiveness of the proposed SAR-ICP method. The information entropy of the reconstructed 3D image using the SAR-ICP is approximately 0.05 times that of the conventional SAR method, indicating a superior image quality. In the future, we anticipate the widespread application of this method in areas such as security inspection, non-destructive testing, and other complex scenarios.

Optical multiple-image authentication method based on Fourier single-pixel imaging and multiple logistic maps.

As a promising technique, the spatial information of an object can be acquired by employing active illumination of sinusoidal patterns in the Fourier single-pixel imaging. However, the major challenge in this field is that a large number of illumination patterns should be generated to record measurements in order to avoid the loss of object details. In this paper, an optical multiple-image authentication method is proposed based on sparse sampling and multiple logistic maps. To improve the measurement efficiency, object images to be authenticated are randomly sampled based on the spatial frequency distribution with smaller size, and the Fourier sinusoid patterns generated for each frequency are converted into binarized illumination patterns using the Floyd-Steinberg error diffusion dithering algorithm. In the generation process of the ciphertext, two chaotic sequences are used to randomly select spatial frequency for each object image and scramble all measurements, respectively. Considering initial values and bifurcation parameters of logistic maps as secret keys, the security of the cryptosystem can be greatly enhanced. For the first time to our knowledge, how to authenticate the reconstructed object image is implemented using a significantly low number of measurements (i.e., at a very low sampling ratio less than 5% of Nyquist limit) in the Fourier single-pixel imaging. The experimental results as well as simulations illustrate the feasibility of the proposed multiple-image authentication mechanism, which can provide an effective alternative for the related research.