Search

Great advances in stitching high-quality retinal images have been made in recent years. On the other hand, very few studies have been carried out on low-resolution retinal imaging. This work investigates the challenges of low-resolution retinal images obtained by the D-EYE smartphone-based fundus camera. The proposed method uses homography estimation to register and stitch low-quality retinal images into a cohesive mosaic. First, a Siamese neural network extracts features from a pair of images, after which the correlation of their feature maps is computed. This correlation map is fed through four independent CNNs to estimate the homography parameters, each specializing in different corner coordinates. Our model was trained on a synthetic dataset generated from the Microsoft Common Objects in Context (MSCOCO) dataset; this work added an important data augmentation phase to improve the quality of the model. Then, the same is evaluated on the FIRE retina and D-EYE datasets for performance measurement using the Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM). The obtained results are promising: the average PSNR was 26.14 dB, with an SSIM of 0.96 on the D-EYE dataset. Compared to the method that uses a single neural network for homography calculations, our approach improves the PSNR by 7.96 dB and achieves a 7.86% higher SSIM score. [ABSTRACT FROM AUTHOR]

Lip print recognition technology originated in the field of forensic medicine, and convolutional neural network has made breakthrough achievements in the field of pattern recognition and machine vision. Convolutional neural network (CNN) algorithm is rarely used in lip pattern recognition. Further exploration and research on the network model suitable for lip pattern recognition. Lip print recognition algorithm based on depth convolution neural network aims to solve the problems of complex image preprocessing, difficult feature extraction and low recognition efficiency in traditional lip print recognition algorithms. It includes collecting lip print images to establish data sets, selecting different CNN models to conduct performance evaluation experiments on low resolution lip print data sets, and analyzing the experimental results with model evaluation indicators. [ABSTRACT FROM AUTHOR]

The Scene Graph is a modern structured representation of an image scene that explicitly describes the scene as a set of objects, attributes, and links between the objects (relationships). With the great advancements in the computer vision field, researchers dedicated their efforts towards more complex reasoning and a high level of understanding of visual scenes. Tasks like Visual Question Answering, image generation, and cross-modal retrieval are examples of Complex vision tasks that require a high level of visual scene understanding. Scene Graph is an effective data structure that highlights complex visual relationships presented in a scene. In this work, we provide a comparative analysis of Scene Graph Generation (SGG) backbone models. The contributed work aims to compare the Convolution Neural Networks (CNN) backbones and the vision transformer-based backbones using the RelTR model. The conducted analysis proved that both SwiftFormer L3 and MiT-B2 transformer backbones increased the model performance over the ResNet50 CNN backbone by 2.1 % and 2.5% Recall@50 respectively when experimented on the same Visual Genome 50 test split. The Visual Genome 50 is a tailored version of The Visual Genome dataset. It contains only the 50 most common relationships and the most frequent 150 object classes. [ABSTRACT FROM AUTHOR]

Crystallography at low resolution must determine the atomic model from less experimental observations, which is challenging in the absence of a model. In addition, model bias is more severe when independent experimental data are scarce. Our methods solve the phase problem by combining the location of accurate model fragments using Phaser with density modification and interpretation of the resulting maps using SHELXE. From a partial, correct structure, the density modification process and the stereochemical constraints draw the rest of the structure, validating the result. This same principle is now exploited at low resolution. Coiled coils are important, ubiquitous structures but notoriously difficult to phase and to predict. Both correct solutions and incorrect ones are poorly discriminated by the crystallographic figures of merit as long as helices are correctly oriented. We incorporate coiled‐coil verification, designed to set up competing, incompatible structural hypotheses to probe both the results and establish the power of the data to discriminate them. Efficiency of coiled‐coil phasing and validation in test cases from 3 to 4 Å is demonstrated in ARCIMBOLDO_LITE, placing single helices, and in ARCIMBOLDO_SHREDDER, with fragments derived from AlphaFold models. SHELXE tracing at low resolution has been enhanced, maintaining its local character but extending the environment assessment. For non‐helical structures, verification is demonstrated in the fragment location process. Its use is exemplified with the solution of the VSR1 structure at 3.5 Å, depending on LLG optimization and the emergence of new features in the electron density. Relying on verification, we have extended the use of the ARCIMBOLDO software to low resolution. [ABSTRACT FROM AUTHOR]

Plant diseases pose a significant threat to agricultural productivity, emphasizing the essential need for early detection and diagnosis. While recent deep learning approaches have made strides in plant disease detection, existing methods fall short in accuracy when handling low-resolution (LR) images, often encountered in field conditions due to limited lighting and variable weather. To address this problem, this research introduces a novel Adaptive scale Feature Extraction Super-Resolution Network (ASFESRN) for corn leaf disease detection. The proposed ASFESRN leverages a unique Feature Dismemberment and Integration Block (FDIB) and Adaptive scale Group Convolutional Network (ASGCN) modules, specifically designed to enhance the resolution of LR field images. Unlike previous group convolution methods, the ASGCN module simultaneously extracts spatial and channel information from the original feature map, overcoming limitations and preserving data transmission. The FDIB, incorporating a Channel Attention Module, three ASGCN blocks, and five convolution layers, further enhances the network’s performance in image super resolution. In comparison to existing methods, the ASFESRN demonstrates superior performance in image super resolution, exhibiting improved PSNR, SSIM. The obtained super-resolved (SR) images are then fed into the Disease Detection Block (DDB) for accurate identification of corn leaf diseases. The proposed ASFESRN model underwent extensive training and testing on the PlantVillage Corn Leaf Disease dataset, encompassing various super resolution scaling factors. Results showcase the effectiveness of the proposed model, with accuracies surpassing existing methods for Diverse super resolution scaling factors. The proposed ASFESRN achieves accuracies of 99.7402%, 98.4805%, and 98.9610%, respectively for super resolution scaling factors (희2, 희4, 희6), offering a robust solution for real-time, accurate disease diagnosis in challenging field conditions. This innovative approach equips farmers with early and accurate diagnosis of corn leaf diseases, aiding in effective crop protection. [ABSTRACT FROM AUTHOR]

The cetane number (CN) is an important fuel property to consider for compression ignition engines as it is a measure of a fuel's ignition delay. Derived cetane number (DCN) already varies significantly within jet fuels. With the expected increasing prevalence of alternative jet fuels, additional variability is expected. DCN is usually assigned to fuels using ASTM methods that use large equipment like the ignition quality tester (IQT), which consumes a lot of fuel and is cumbersome to operate. Over the last decade, there have been advances in the development of chemometric models, which use machine learning to correlate infrared spectra of fuels to fuel properties like DCN, density, and C/H ratio, among many others. These techniques have certain advantages over the ASTM methods, and previous studies performed on samples of diesel fuels have shown high accuracies in DCN prediction. However, this accuracy is generally a result of high resolution, making the equipment expensive, relatively large for handheld sensors, and power-hungry. On the other hand, nondispersive infrared (NDIR) sensors, despite having a low resolution, are attractive because they can be compact, inexpensive, and power efficient. These characteristics are important for handheld or onboard fuel sensors. However, one would anticipate a tradeoff between these advantages and accuracy. This study investigates this tradeoff and the feasibility of low-resolution NDIR sensors to discern fuel properties such as DCN by using Machine Learning models trained on real FTIR data, and DCNs obtained from IQT. DCN predictions were made for blends of ATJ/F-24, CN fuels, and neat Jet A1, A2, and JP 5, with an error limit of 10%. It was found that there seems to be sufficient variability in the near infrared (NIR) range to discern DCN with a feasible number of channels, but the channels have to be narrow (e.g., FWHMs as narrow as 60 nm). For the dataset in the study, the performance of linear models was better than the nonlinear model. Finally, NIR region beyond 1050 nm was found to be more important in DCN prediction, primarily the regions consisting of the first and second C-H overtones and the C-H combination band. [ABSTRACT FROM AUTHOR]

While comparing to the additional medical imaging modalities, the low resolution (LR) with poor quality images are obtained because of natural intrinsic imaging characteristics. We proposed a novel deep learning- based super-resolution of ultrasound images that are texture-driven. In this study, Convolutional Neural Networks (CNNs) are used to speed up the process to increase the image quality. Additionally, the Dwarf Mongoose Optimization (DMO) method is used to adjust the parameters of CNN thereby improving the quality of the image resolutions. The textures of the super-resolution images are examined with the Histogram of Oriented Gradients (HOG). The experimental works are handled using python software. Super-resolution per second and PSNR/dB values for the proposed model are 0.289 and 42.340, respectively. This model also offers FSO, GMSD, MAD, and MSSIM values of 0.9712, 0.013, 0.9802, and 0.9832 which is relatively higher than the existing techniques. [ABSTRACT FROM AUTHOR]

Arnica montana L. is a medicinal plant with significant conservation importance. It is crucial to monitor this species, ensuring its sustainable harvesting and management. The aim of this study is to develop a practical system that can effectively detect A. montana inflorescences utilizing unmanned aerial vehicles (UAVs) with RGB sensors (red–green–blue, visible light) to improve the monitoring of A. montana habitats during the harvest season. From a methodological point of view, a model was developed based on a convolutional neural network (CNN) ResNet101 architecture. The trained model offers quantitative and qualitative assessments of A. montana inflorescences detected in semi-natural grasslands using low-resolution imagery, with a correctable error rate. The developed prototype is applicable in monitoring a larger area in a short time by flying at a higher altitude, implicitly capturing lower-resolution images. Despite the challenges posed by shadow effects, fluctuating ground sampling distance (GSD), and overlapping vegetation, this approach revealed encouraging outcomes, particularly when the GSD value was less than 0.45 cm. This research highlights the importance of low-resolution image clarity, on the training data by the phenophase, and of the need for training across different photoperiods to enhance model flexibility. This innovative approach provides guidelines for mission planning in support of reaching sustainable management goals. The robustness of the model can be attributed to the fact that it has been trained with real-world imagery of semi-natural grassland, making it practical for fieldwork with accessible portable devices. This study confirms the potential of ResNet CNN models to transfer learning to new plant communities, contributing to the broader effort of using high-resolution RGB sensors, UAVs, and machine-learning technologies for sustainable management and biodiversity conservation. [ABSTRACT FROM AUTHOR]

Eye tracking is prevalent in scientific and commercial applications. Recent computer vision and deep learning methods enable eye tracking with off-the-shelf webcams and reduce dependence on expensive, restrictive hardware. However, such deep learning methods have not yet been applied and evaluated for remote, online psychological experiments. In this study, we tackle critical challenges faced in remote eye tracking setups and systematically evaluate appearance-based deep learning methods of gaze tracking and blink detection. From their own homes and laptops, 65 participants performed a battery of eye tracking tasks including (i) fixation, (ii) zone classification, (iii) free viewing, (iv) smooth pursuit, and (v) blink detection. Webcam recordings of the participants performing these tasks were processed offline through appearance-based models of gaze and blink detection. The task battery required different eye movements that characterized gaze and blink prediction accuracy over a comprehensive list of measures. We find the best gaze accuracy to be 2.4° and precision of 0.47°, which outperforms previous online eye tracking studies and reduces the gap between laboratory-based and online eye tracking performance. We release the experiment template, recorded data, and analysis code with the motivation to escalate affordable, accessible, and scalable eye tracking that has the potential to accelerate research in the fields of psychological science, cognitive neuroscience, user experience design, and human–computer interfaces. [ABSTRACT FROM AUTHOR]

Gaze estimation significantly enhances user interfaces, road safety, accessibility for the disabled, and consumer behavior analysis. Traditional methods depend on high-resolution images and ideal conditions, neglecting low-resolution scenarios. We propose GAFUSE-Net, designed for low-resolution gaze estimation. GAFUSE-Net includes a novel Facial Attention-Enhanced feature extraction (FAE) module and a Local image Super-resolution based Eye feature extraction (LSE) module for low resolution gaze estimation. The FAE module extracts gaze-relevant features from facial images. The LSE module uses super-resolution to enhance low-resolution eye images, followed by feature extraction with DeepEyeNet. Combined facial and eye features enable accurate gaze estimation. Experiments on the MPIIFaceGaze dataset show GAFUSE-Net’s effectiveness, achieving mean angular errors of 4.37°, 4.62°, 4.84°, and 6.19° at resolutions of $128\times 128$ , $64\times 64$ , $32\times 32$ , and $16\times 16$ , respectively. These errors are 0.2%, 1.3%, 5.7%, and 10.2% lower than state-of-the-art methods. The experimental results highlight the efficiency of GAFUSE-Net across various resolutions, significantly advancing the practical implementation of gaze estimation.

Great advances in stitching high-quality retinal images have been made in recent years. On the other hand, very few studies have been carried out on low-resolution retinal imaging. This work investigates the challenges of low-resolution retinal images obtained by the D-EYE smartphone-based fundus camera. The proposed method uses homography estimation to register and stitch low-quality retinal images into a cohesive mosaic. First, a Siamese neural network extracts features from a pair of images, after which the correlation of their feature maps is computed. This correlation map is fed through four independent CNNs to estimate the homography parameters, each specializing in different corner coordinates. Our model was trained on a synthetic dataset generated from the Microsoft Common Objects in Context (MSCOCO) dataset; this work added an important data augmentation phase to improve the quality of the model. Then, the same is evaluated on the FIRE retina and D-EYE datasets for performance measurement using the Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM). The obtained results are promising: the average PSNR was 26.14 dB, with an SSIM of 0.96 on the D-EYE dataset. Compared to the method that uses a single neural network for homography calculations, our approach improves the PSNR by 7.96 dB and achieves a 7.86% higher SSIM score.

Arnica montana L. is a medicinal plant with significant conservation importance. It is crucial to monitor this species, ensuring its sustainable harvesting and management. The aim of this study is to develop a practical system that can effectively detect A. montana inflorescences utilizing unmanned aerial vehicles (UAVs) with RGB sensors (red–green–blue, visible light) to improve the monitoring of A. montana habitats during the harvest season. From a methodological point of view, a model was developed based on a convolutional neural network (CNN) ResNet101 architecture. The trained model offers quantitative and qualitative assessments of A. montana inflorescences detected in semi-natural grasslands using low-resolution imagery, with a correctable error rate. The developed prototype is applicable in monitoring a larger area in a short time by flying at a higher altitude, implicitly capturing lower-resolution images. Despite the challenges posed by shadow effects, fluctuating ground sampling distance (GSD), and overlapping vegetation, this approach revealed encouraging outcomes, particularly when the GSD value was less than 0.45 cm. This research highlights the importance of low-resolution image clarity, on the training data by the phenophase, and of the need for training across different photoperiods to enhance model flexibility. This innovative approach provides guidelines for mission planning in support of reaching sustainable management goals. The robustness of the model can be attributed to the fact that it has been trained with real-world imagery of semi-natural grassland, making it practical for fieldwork with accessible portable devices. This study confirms the potential of ResNet CNN models to transfer learning to new plant communities, contributing to the broader effort of using high-resolution RGB sensors, UAVs, and machine-learning technologies for sustainable management and biodiversity conservation.

High Resolution (HR) images have numerous applications, such as video conferencing, remote sensing, medical imaging, etc. Furthermore, a few challenges with the super resolution algorithms of magnetic resonance brain images are now obtainable, namely, low sensitivity, significant frequency noise as well as poor resolution. To fix these problems, a Convolutional Neural Network (CNN) based Discrete Cosine Transform (DCT) singular frame quality improvement method is described. There are two stages in this proposed method, involving training and testing. During the training stage, the HR, and Low Resolution (LR) pictures are employed as input, and they are preprocessed to create blocks of images. The histogram and DCT are used for extracting the features from the LR and HR blocks, and these extracted features are assigned with class id. The CNN, which extracts the features and allocates class id, receives its feature extractor as its final input. An LR input image is once more divided into [2 × 2] blocks during the testing stage, so each block histogram and DCT feature are estimated. Each feature vector is fed into the neural network as well as the results are contrasted with a set of feature vectors that have been recorded, in addition to the class id that has been allocated to a certain vector. In order to generate a Super resolution image with an LR image, a relevant HR block is then swapped out for this LR block. These results indicated that the initial dataset can achieve 22.4 and 19.5 Peak Signal to Noise Ratio (PSNR) and Root Mean Square Error (RMSE) values while measuring the effectiveness of this proposed method using RMSE and PSNR. Then, the second dataset illustrates that the PSNR and RMSE values are 20.1 and 25.5. For the third dataset, the values are 45.7 and 12.3, respectively. However, the presented method works better than the neural method of Super Resolution Channel Spatial Modulation Network and resolution enhancement technique.

Plant diseases significantly impact agricultural productivity. Early disease identification and diagnosis are critical for plant protection. Recent deep learning approaches have substantially aided detection of plant diseases. However, existing plant leaf disease identification techniques do not provide sufficient disease detection accuracies when the input image resolution is low. Images of grape leaves taken in the field may be low-resolution (LR) in nature due to limited lighting and varying weather conditions. Such LR images may result in incorrect real-time disease diagnosis. Hence super-resolution is a way to solve this problem. This research work proposes a novel Residual Skip Network-based Super-Resolution for Leaf Disease Detection (RSNSR-LDD) in the Grape plant. The input LR image is separated into two subcomponents using the guided filtering technique. Features from the two subcomponents are extracted using single convolutional layer and four layers of the two-channel residual skip network. Concatenation is used to combine the features of two channels. Following this, a decoding block and a convolutional layer are utilized to generate the super-resolution (SR) image. A new collaborative loss function is proposed for training. The obtained SR image is given to the Disease Detection Network (DDN) for grape leaf disease detection. This approach, based on an LR image, assists the farmer in spotting grape plant diseases very early. Proposed model was extensively trained and tested on PlantVillage, Grape 400, Grape Leaf Disease datasets with multiple super-resolution scaling factors for various grape leaves images. For different super-resolution scaling factors such as Χ2, Χ4, Χ6, the proposed model RSNSR-LDD achieved accuracies of 97.19%, 99.37%, and 99.06% for the PlantVillage dataset, 96.88%, 97.12%, and 95.43% for the Grape400 dataset, and 100% for the Grape Leaf Disease dataset. [ABSTRACT FROM AUTHOR]

Facial expression recognition (FER) in the wild has attracted much attention in recent years due to its wide range of applications. Most current approaches use deep learning models trained on relatively large images, which significantly reduces their accuracy when they have to infer low-resolution images. In this paper, a residual voting network is proposed for the classification of low-resolution facial expression images. Specifically, the network consists of a modified ResNet-18, which divides each sample into multiple overlapping crops, makes a prediction of the class to which each of the crops belongs, and by soft-voting the predictions of all the crops, the network determines the class of the sample. A novel aspect of this work is that the image splitting is not performed before entering the network, but at an intermediate point in the network, which significantly reduces the resource consumption. The proposed approach was evaluated on two popular benchmark datasets (AffectNet and RAF-DB) by scaling the images to a network input size of 48 × 48. The proposed model reported an accuracy of 63.06% on AffectNet and 85.69% on RAF-DB with seven classes in both cases, which are values comparable to those provided by other current approaches using much larger images. [ABSTRACT FROM AUTHOR]

The physics of Terahertz (THz) technology is the electromagnetic (EM) spectrum band between the infrared and the microwave band with frequencies of about 0.1 to 30 THz. THz signals have gained adoption in medicine, telecommunications, security monitoring and imaging. THz imaging technology has the advantages of rapid imaging, strong penetration, and harmless to the human body hence widely used in a variety of security environments and has become an alternative technology for X-ray imaging. However, THz is characterized by low resolution of THz images of which noise is an integral factor constituting a defect. Clarity of THz image is therefore essential at various security checkpoints to avoid life’s dangers and treats. In this paper, we propose an efficient and high-performance defect detection model based on RetinaNet to recognize defects from captured images. The strategy of transfer learning is introduced to improve detection performance accuracy, which enhances the average precision (AP) by 19.2%. Contrary to existing THz image detection techniques on image recognition pertaining to the whole region of the image, we adopt a different approach via differential evolution search algorithm for optimization given the small proportion of defect area which improves the AP by 9.9% comparing with the fine-tuned model. For the problem of the lack of defect data samples, image augmentation is adopted to enrich our training samples, which improves the AP by 9.5%. As for the problem of low precision and recall in detecting blurred images, we firstly manually generate clear-blurred image pairs to train a GAN. Then, the blurred images are deblurred using a trained generator. We get 5.5% AP improvement on the testset using our approach. Compared with existing works, the optimized model based on RetinaNet has better detection performance, subsequently, proving the practicability and effectiveness of the proposed method.

Crime prediction in video-surveillance systems is required to prevent incident and protect assets. In this sense, our article proposes first artificial intelligence approach for Robbery Behavior Potential (RBP) prediction and detection in an indoor camera. Our method is based on three detection modules including head cover, crowd and loitering detection modules for timely actions and preventing robbery. The two first modules are implemented by retraining YOLOV5 model with our gathered dataset which is annotated manually. In addition, we innovate a novel definition for loitering detection module which is based on DeepSORT algorithm. A fuzzy inference machine renders an expert knowledge as rules and then makes final decision about predicted robbery potential. This is laborious due to: different manner of robber, different angle of surveillance camera and low resolution of video images. We accomplished our experiment on real world video surveillance images and reaching the F1-score of 0.537. Hence, to make an experimental comparison with the other related works, we define threshold value for RBP to evaluate video images as a robbery detection problem. Under this assumption, the experimental results show that the proposed method performs significantly better in detecting the robbery as compared to the robbery detection methods by distinctly report with F1-score of 0.607. We strongly believe that the application of the proposed method could cause reduction of robbery detriment in a control center of surveillance cameras by predicting and preventing incident of robbery. On the other hand, situational awareness of human operator enhances and more cameras can be managed.

This article presents a conversation with Laura U. Marks, a distinguished scholar of film and new media who launched the Small File Media Festival in 2020. Marks' message is that large-file streaming is harmful to the planet and that we need to act against its growing yet under-the-radar carbon footprint. The festival's goal is to demonstrate that movies can be captivating and immersive even if they are not large in file size, and encourages creators to experiment with low-energy technologies and techniques. Given the high stakes, the festival's organizers hope that small-file media will become increasingly influential as an aesthetic movement. [ABSTRACT FROM AUTHOR]

The three-dimensional perception applications have been growing since Light Detection and Ranging devices have become more affordable. On those applications, the navigation and collision avoidance systems stand out for their importance in autonomous vehicles, which are drawing an appreciable amount of attention these days. The on-road object classification task on three-dimensional information is a solid base for an autonomous vehicle perception system, where the analysis of the captured information has some factors that make this task challenging. On these applications, objects are represented only on one side, its shapes are highly variable and occlusions are commonly presented. But the highest challenge comes with the low resolution, which leads to a significant performance dropping on classification methods. While most of the classification architectures tend to get bigger to obtain deeper features, we explore the opposite side contributing to the implementation of low-cost mobile platforms that could use low-resolution detection and ranging devices. In this paper, we propose an approach for on-road objects classification on extremely low-resolution conditions. It uses directly three-dimensional point clouds as sequences on a transformer-convolutional architecture that could be useful on embedded devices. Our proposal shows an accuracy that reaches the 89.74 % tested on objects represented with only 16 points extracted from the Waymo, Lyft's level 5 and Kitti datasets. It reaches a real time implementation (22 Hz) in a single core processor of 2.3 Ghz.

Image upscaling employs various polynomial techniques, including bicubic and Lanczos, to minimize computing load for various real-time applications. Polynomial interpolation, on the other hand, results in blurring artifacts in high resolution (HR) images due to edge degradation. Many other interpolation approaches, including edge-based and dictionary learning-based schemes, result in blurring in the image's high frequency (HF) region. As a result, a novel post-processing technique is proposed that uses the Lanczos method to upscale the low resolution (LR) image. To eliminate blur along the image's edge, the edge component of the interpolated image is retrieved. Adaptive sharpening is used to make the edges more prominent, as predicting pixels in the high variance region is challenging. After the edge is extracted, the smooth region is filtered with an optimized directional anisotropic diffusion (ODAD) filter to preserve texture details and geometric regularity. The sharpened edge image is aggregated with the ODAD filtered image to generate an improved HR (IHR) image. The IHR is then scaled up and down to reduce blurring even further, resulting in a blurred HR image (BHR). The residual image is obtained by subtracting the BHR from the IHR images, which represent the undetected HF. Finally, the residual image and the IHR image are combined to generate the final restored image, which includes all prediction loss details. The results of the experiments show that the proposed algorithm outperforms existing methodologies, particularly in terms of visuality and parameter-wise measurement. • A novel post-processing upscaling technique is used to increase the visual quality of the upscaled image. • An AES and an ODAD filter extract very HF and fine texture details. • The residual image is utilized to restore more fine details to the HR image. [ABSTRACT FROM AUTHOR]

It has been widely noticed the performance of algorithms for high-resolution face recognition (HRFR) degrades significantly for low-resolution face recognition (LRFR). In this paper, we discover the main source of this performance degradation comes from the human-defined inductive bias of CNN, which constrains the model to absorb effective information and leads to overfitting. To overcome the shortcoming, for the first time, we adopt a transformer-based DNN algorithm named DeiT to accomplish LRFR tasks. On the other hand, we further borrow the form of knowledge distillation. Traditional knowledge distillation network for LRFR sets the student model to be simpler than or the same as the teacher model while using the teacher model off-the-shelf, leading to the model capacity gap. Instead, we fuse an online network into the original parameter-fixed teacher model to learn how to transfer knowledge. The final "knowledge" is the sum fusion of the outputs of both the teacher model and the student model. Experiments show that even without training on LR face images, the performance of our model can be comparable to recent baselines delicately designed for LRFR tasks on certain LR face datasets. After the entire training process, in either real-world LR datasets, artificially down-sampled datasets, or generated LR face datasets, our method performs favorably against the state-of-the-art (SOTA) algorithms. [ABSTRACT FROM AUTHOR]

Crystallography at low resolution must determine the atomic model from less experimental observations, which is challenging in the absence of a model. In addition, model bias is more severe when independent experimental data are scarce. Our methods solve the phase problem by combining the location of accurate model fragments using Phaser with density modification and interpretation of the resulting maps using SHELXE. From a partial, correct structure, the density modification process and the stereochemical constraints draw the rest of the structure, validating the result. This same principle is now exploited at low resolution. Coiled coils are important, ubiquitous structures but notoriously difficult to phase and to predict. Both correct solutions and incorrect ones are poorly discriminated by the crystallographic figures of merit as long as helices are correctly oriented. We incorporate coiled-coil verification, designed to set up competing, incompatible structural hypotheses to probe both the results and establish the power of the data to discriminate them. Efficiency of coiled-coil phasing and validation in test cases from 3 to 4 Å is demonstrated in ARCIMBOLDO_LITE, placing single helices, and in ARCIMBOLDO_SHREDDER, with fragments derived from AlphaFold models. SHELXE tracing at low resolution has been enhanced, maintaining its local character but extending the environment assessment. For non-helical structures, verification is demonstrated in the fragment location process. Its use is exemplified with the solution of the VSR1 structure at 3.5 Å, depending on LLG optimization and the emergence of new features in the electron density. Relying on verification, we have extended the use of the ARCIMBOLDO software to low resolution., (© 2024 The Author(s). Protein Science published by Wiley Periodicals LLC on behalf of The Protein Society.)

This study proposes a simple convolutional neural network (CNN)-based model for vehicle classification in low resolution surveillance images collected by a standard security camera installed distant from a traffic scene. In order to evaluate its effectiveness, the proposed model is tested on a new dataset containing tiny (100 × 100 pixels) and low resolution (96 dpi) vehicle images. The proposed model is then compared with well-known VGG16-based CNN models in terms of accuracy and complexity. Results indicate that although the well-known models provide higher accuracy, the proposed method offers an acceptable accuracy (92.9%) as well as a simple and lightweight solution for vehicle classification in low quality images. Thus, it is believed that this study might provide useful perception and understanding for further research on the use of standard low-cost cameras to enhance the ability of the intelligent systems such as intelligent transportation system applications. [ABSTRACT FROM AUTHOR]

In the commonly used datasets of person re-identification, the image quality is not uniform. Most existing methods on person re-identification mainly focus on the challenges caused by occlusion, view and pose variations, ignoring the diversity of person image quality. In this paper, we provide an intuitive solution to address this problem. Specifically, we generate low-resolution images by reducing the resolution of original person images and propose a low-resolution assisted three-stream network (LRAN) to fuse the extracted person features from original RGB images, low-resolution images and greyscale images into a more robust feature as the final person representation. In this way, the model eliminates the impact of image quality differences to some extent. Experimental results demonstrate that the proposed method achieves the state-of-the-art results on Market-1501, DukeMTMC-reID and CUHK03-NP datasets. [ABSTRACT FROM AUTHOR]

Copyright of Savoirs En Prisme is the property of Universitaires de Reims and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

With increasing usage of face recognition algorithms, it is well established that external artifacts and makeup accessories can be applied to different facial features such as eyes, nose, mouth, and cheek, to obfuscate one’s identity or to impersonate someone else’s identity. Recognizing faces in the presence of these artifacts comprises the problem of disguised face recognition, which is one of the most arduous covariates of face recognition. The challenge becomes exacerbated when disguised faces are captured in real-time environment, with low resolution images. To address the challenge of disguised face recognition, this paper first proposes a novel multi-objective encoder-decoder network, termed as DED-Net. DED-Net attempts to learn the class variations in the feature space generated by both disguised as well non-disguised images, using a combination of Mahalanobis and Cosine distance metrics, along with Mutual Information based supervision. The DED-Net is then extended to learn from the local and global features of both disguised and non-disguised face images for efficient face recognition, and the complete framework is termed as Disguise Resilient (D-Res) framework. The efficacy of the proposed framework has been demonstrated on two real-world benchmark datasets: Disguised Faces in the Wild (DFW) 2018 and DFW2019 competition datasets. In addition, this research also emphasizes on the importance of recognizing disguised faces in low resolution settings and proposes three experimental protocols to simulate the real-world surveillance scenario. To this effect, benchmark results have been shown on seven protocols for three low resolution settings ($32\times 32$ , $24\times 24$ , and $16\times 16$) of the two DFW benchmark datasets. The results demonstrate superior performance of the D-Res framework, in comparison with benchmark algorithms. For example, an improvement of around 3% is observed on the Overall protocol of the DFW2019 dataset, where the D-Res framework achieves 96.3%. Experiments have also been performed on benchmark face verification datasets (LFW, YTF, and IJB-B), where the D-Res framework achieves improved verification accuracy. [ABSTRACT FROM AUTHOR]

Current face recognition tasks are usually carried out on high-quality face images, but in reality, most face images are captured under unconstrained or poor conditions, e.g., by video surveillance. Existing methods are featured by learning data uncertainty to avoid overfitting the noise, or by adding margins to the angle or cosine space of the normalized softmax loss to penalize the target logit, which enforces intra-class compactness and inter-class discrepancy. In this paper, we propose a deep Rival Penalized Competitive Learning (RPCL) for deep face recognition in low-resolution (LR) images. Inspired by the idea of the RPCL, our method further enforces regulation on the rival logit, which is defined as the largest non-target logit for an input image. Different from existing methods that only consider penalization on the target logit, our method not only strengthens the learning towards the target label, but also enforces a reverse direction, i.e., becoming de-learning, away from the rival label. Comprehensive experiments demonstrate that our method improves the existing state-of-the-art methods to be very robust for LR face recognition. [ABSTRACT FROM AUTHOR]

Fine-grained object classification (FGOC) focuses on identifying subcategories of objects, which is crucial in military and civilian. Existing FGOC methods primarily focus on high-resolution aerial images, limiting their application on low-resolution (LR) FGOC that is a more realistic setting, especially on resource-constrained satellite devices. It is more challenging to deal with LR FGOC since objects’ details are blurred or missing. Addressing this issue, we make the first attempt to explore LR FGOC and propose a novel pipeline based on two technical insights: 1) feature balance strategy discriminatively integrates super-resolution weak and strong detailed presentations into coarse features of LR aerial images, achieving a feature balance to avoid that the weak detailed presentations are inhibited by the strong ones and 2) iterative interaction mechanism alternately refines feature details of the discriminative ship regions and optimizes the performance of FGOC. Moreover, we build a low-resolution fine-grained object (LFS) dataset to promote further study and evaluation. Extensive experiments on the proposed LFS dataset and the other three object datasets of DOTA, FS23, and HRSC2016 demonstrate that our method outperforms state-of-the-art algorithms. Dataset and code are publicly available at https://github.com/wdzhao123/FBNet. [ABSTRACT FROM AUTHOR]

Face recognition can be installed in a surveillance system so that it can be used for monitoring, tracking and access control. An excellent, intelligent surveillance system should be sensitive to the objects far away from the camera. Unfortunately, due to the long-distance, objects like human faces captured by the camera are too small to identify. As to enhance the subtle color differences in the face image, in this paper we first improve the resolution of the captured image using deep convolution neural networks (DCNNs). Then the efficient features are extracted and used to do classification. As for verifying the effectiveness of the proposed method, we used three databases including AR face database, Georgia Tech face database (GT) database, and Labelled Faces in the Wild (LFW) database, altogether, to conduct the training and testing. Compared to the existing approaches, experimental results show that the identification accuracy of the proposed method outperforms any existing approaches. [ABSTRACT FROM AUTHOR]

Human action recognition is one of the most pressing questions in societal emergencies of any kind. Technology is helping to solve such problems at the cost of stealing human privacy. Several approaches have considered the relevance of privacy in the pervasive process of observing people. New algorithms have been proposed to deal with low-resolution images hiding people identity. However, many of these methods do not consider that social security asks for real-time solutions: active cameras require flexible distributed systems in sensible areas as airports, hospitals, stations, squares and roads. To conjugate both human privacy and real-time supervision, we propose a novel deep architecture, the Multi Streams Network. This model works in real-time and performs action recognition on extremely low-resolution videos, exploiting three sources of information: RGB images, optical flow and slack mask data. Experiments on two datasets show that our architecture improves the recognition accuracy compared to the two-streams approach and ensure real-time execution on Edge TPU (Tensor Processing Unit).

Copyright of Journal of Signal Processing is the property of Journal of Signal Processing and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Due to the limitation of the pixel size in the focal plane, the low-resolution(LR) infrared sensor has a very low image resolution when sampling and imaging scenes with slightly rich spatial frequencies, and aliasing is sometimes very serious. This paper uses a new technique base on sub-pixel displacement to reconstruct high-resolution(HR) images,with reduced aliasing, from a sequence of under-sampled rotated frames of the same object. First, this paper presents an image degradation model, based on the under-sampling model of the infrared image and the infrared radiation distribution on the focal plane. Second, an image reconstruction algorithm based on image micro rotation is proposed and implemented to solve the problems of inaccurate temperature measurement and target recognition caused by low resolution. Finally, the experiments results are provided to test our algorithm, and we can obtain the image whose resolution is four or five times higher than the under-sampled frames, as well as improve the temperature measurement accuracy by more than 10%. The experimental results also show that the image reconstruction algorithm is very robust, efficient and has a good reconstruction effect.

Abstract The proficiency of image processing is of extreme importance in perceiving and collecting information from the images, which includes the process of changing or interpreting existing images. In medical image processing, imaging with more accuracy plays a crucial role in better diagnosis or for the posterior analysis of treatment. Magnetic Resonance Imaging (MRI) is a medicinal creative tool for studying the internal structures and functionalities of human brain, knee, heart, liver, etc. Typical MR scans are essential now for better diagnosis but, limited resolution that is often inadequate for extracting detailed and reliable information. So, for the super resolution (SR) of MR brain images concepts of compressive sensing (CS) & fuzzy logical rules to improve data quality are proposed in this paper. Usually, reconstruction of an SR image is the formation of high resolution (HR) image which is obtained from one or few low resolution (LR) images. In the proposed method, with the help of compressive sensing a very limited number of images are considered even though it’s a challenging task and fuzzy logical rules for a specific membership function are applied to improve the resolution of the image. To assess the performance of the proposal, different metrics are evaluated and achieved better results.

Depth edge represents contours of depth images. However, captured images often have low resolution. Furthermore, conventional edge detection methods fail to handle depth images with low resolution. To cater depth images with low resolution, we propose a novel edge detection method with high accuracy and computational speed. The proposed method modifies the local binary pattern and selects an adaptive threshold to obtain a rough edge map. Then, it removes irrelevant edges to obtain the result. Experimental results show that the proposed method can rapidly acquire accurate depth edge maps and is applicable to depth images with low resolution. The proposed method is compared with and outperforms other renowned edge detection methods.

This study proposes a simple convolutional neural network (CNN)-based model for vehicle classification in low resolution surveillance images collected by a standard security camera installed distant from a traffic scene. In order to evaluate its effectiveness, the proposed model is tested on a new dataset containing tiny (100 × 100 pixels) and low resolution (96 dpi) vehicle images. The proposed model is then compared with well-known VGG16-based CNN models in terms of accuracy and complexity. Results indicate that although the well-known models provide higher accuracy, the proposed method offers an acceptable accuracy (92.9%) as well as a simple and lightweight solution for vehicle classification in low quality images. Thus, it is believed that this study might provide useful perception and understanding for further research on the use of standard low-cost cameras to enhance the ability of the intelligent systems such as intelligent transportation system applications.