Web Image Research Articles

Object detection analysis based on Machine Learning Algorithms

The branch of artificial intelligence known as computer vision focuses on developing and utilizing digital systems to process, examine, and interpret visual data. Computer vision aims to make it possible for computing equipment to accurately recognise an object's digital image and perform the necessary action.The field of transportation and artificial intelligence includes vehicle discovery, which is pivotal. Intelligent transport system (ITS) is a fashion used to produce an effective road transportation system. The Automated Driving System which has made significant advancements in recent times, heavily relies on vehicle identification in its armature. The fashion of detecting vehicles on the road is used for objects similar as vehicle shadowing, counts, and average speed of each vehicle, business analysis, and vehicle categorization and can be applied in a variety of settings. A wide, active, and grueling field of computer vision is real- time object discovery. This recognizes semantic objects of a particular class in digital prints and pictures. For face recognition, vehicle recognition, rambler counting, web image analysis, security systems, and tone- driving motorcars, object discovery is constantly employed. Key Words: Vehicle dataset, ITS, Vehicle detection, Highway management

Boosting Robust Learning Via Leveraging Reusable Samples in Noisy Web Data

Webly-supervised fine-grained visual classification (FGVC) has attracted increasing attention in recent years because of the unaffordable cost of obtaining correctly-labeled large-scale fine-grained datasets. However, due to the existence of label noise in web images and the high memorization capacity of deep neural networks, training deep fine-grained (FG) models directly through web images tends to have an inferior recognition ability. In the literature, to alleviate this issue, loss correction methods try to estimate the noise transition matrix, but the inevitable false correction would cause accumulated errors. Sample selection methods identify clean (‘`easy’') samples based on the fact that small losses can alleviate the accumulated errors. However, ‘`hard’' and mislabeled examples that can both boost the robustness of FG models are also dropped. To this end, we propose a certainty-based reusable sample selection and correction approach, termed as CRSSC, for coping with label noise in training deep FG models with web images. Our key idea is to additionally identify and correct reusable samples, and then leverage them together with clean examples to update the network. Furthermore, in order to endow our model with the capability to capture richer and more discriminative feature representations, we propose a cross-layer attention-based feature refinement (CLAR) block. We demonstrate the superiority of the proposed approach from both theoretical and experimental perspectives. The source code, models, and data have been made available at https://github.com/NUST-Machine-Intelligence-Laboratory/CLAR-CRSSC.

Guided by Meta-Set: A Data-Driven Method for Fine-Grained Visual Recognition

The lack of sufficient training data has been one obstacle to fine-grained visual classification research because labeling subcategories generally requires specialist knowledge. As one optional approach to alleviating the data-hunger problem, leveraging web images as training data is drawing increasing attention. Nevertheless, web images potentially have false labels, which can misguide the training process. Although several works have been proposed to deal with label noise, it still can be difficult for the network to tackle complex real-world noisy labels without any prior knowledge. In the literature, we propose to leverage a small and clean meta-set to provide reliable prior knowledge for tackling noisy web images. Specifically, our method trains a network with two peer predicting heads, which learn from noisy web images (web head) and meta ones (meta head), respectively. The meta head produces pseudo soft labels for web images to revise their training loss, which can overcome the high noise ratio problem. Furthermore, a selection net is trained in a meta-learning strategy to identify in- and out-of-distribution noisy images. Then in-distribution ones are reused for training with pseudo soft labels produced by the meta head as supervision, while out-of-distribution ones are discarded. In this manner, the misguidance caused by label noise is remarkably alleviated and in-distribution noisy samples are properly exploited to boost model performance. The superiority of our proposed approach is demonstrated by mathematical theory with great interpretability as well as extensive experimental results on the real-world dataset WebFG-496. Our source code is available at:.

The Ionizing Spectra of Extremely Metal-poor O Stars: Constraints from the Only H ii Region in Leo P

Metal-poor, star-forming dwarf galaxies produce extreme nebular emission and likely played a major role in cosmic reionization. Yet, determining their contribution to the high-redshift ionizing photon budget is hampered by the lack of observations constraining the ionizing spectra of individual massive stars more metal-poor than the Magellanic Clouds (20%–50% Z ⊙). We present new Keck Cosmic Web Imager (KCWI) optical integral field unit spectroscopy of the only H ii region in Leo P (3% Z ⊙), which is powered by a single O star. We calculate the required production rate of photons capable of ionizing hydrogen and helium from the observed Hβ and He i λ4471 emission-line fluxes. Remarkably, we find that the ionizing photon production rate and spectral hardness predicted by a tlusty model fit to the stellar spectral energy distribution agrees with our observational measurements within the uncertainties. We then fit Cloudy photoionization models to the full suite of optical emission lines in the KCWI data and show that the shape of the same tlusty ionizing continuum simultaneously matches lines across a wide range of ionization energies. Finally, we detect O iii] and N iii] nebular emission in the Hubble Space Telescope far-ultraviolet spectrum of the Leo P H ii region, and highlight that the rarely observed N iii] emission cannot be explained by our Cloudy models. These results provide the first observational evidence that widely used, yet purely theoretical, model spectra accurately predict the ionizing photon production rate from late-O stars at very low metallicity, validating their use to model metal-poor galaxies both locally and at high redshift.

Secure Content Based Image Retrieval Scheme Based on Deep Hashing and Searchable Encryption

To solve the problem that the existing ciphertext domain image retrieval system is challenging to balance security, retrieval efficiency, and retrieval accuracy. This research suggests a searchable encryption and deep hashing-based secure image retrieval technique that extracts more expressive image features and constructs a secure, searchable encryption scheme. First, a deep learning framework based on residual network and transfer learning model is designed to extract more representative image deep features. Secondly, the central similarity is used to quantify and construct the deep hash sequence of features. The Paillier homomorphic encryption encrypts the deep hash sequence to build a high-security and low-complexity searchable index. Finally, according to the additive homomorphic property of Paillier homomorphic encryption, a similarity measurement method suitable for computing in the retrieval system’s security is ensured by the encrypted domain. The experimental results, which were obtained on Web Image Database from the National University of Singapore (NUS-WIDE), Microsoft Common Objects in Context (MS COCO), and ImageNet data sets, demonstrate the system’s robust security and precise retrieval, the proposed scheme can achieve efficient image retrieval without revealing user privacy. The retrieval accuracy is improved by at least 37% compared to traditional hashing schemes. At the same time, the retrieval time is saved by at least 9.7% compared to the latest deep hashing schemes.

Cold mode gas accretion on two galaxy groups at z ∼ 2

ABSTRACT We present Keck Cosmic Web Imager (KCWI) integral field spectroscopy (IFS) observations of rest-frame UV emission lines $\rm Ly\alpha$, CIVλλ 1548 Å, 1550Å, and $\rm HeII$ 1640 Å observed in the circumgalactic medium (CGM) of two z = 2 radio-loud quasar host galaxies. We detect extended emission on 80–90 kpc scale in $\rm Ly\alpha$ in both systems with CIV and $\rm HeII$ emission also detected out to 30–50 kpc. All emission lines show kinematics with a blue and redshifted gradient pattern consistent with velocities seen in massive dark matter haloes and similar to kinematic patterns of inflowing gas seen in hydrodynamical simulations. Using the kinematics of both resolved $\rm Ly\alpha$ emission and absorption, we can confirm that both kinematic structures are associated with accretion. Combining the KCWI data with molecular gas observations with Atacama Large Millimeter/submillimeter Array (ALMA) and high-spatial resolution of ionized gas with Keck OSIRIS, we find that both quasar host galaxies reside in proto-group environments at z = 2. We estimate 1–6 × 1010M⊙ of warm-ionized gas within 30–50 kpc from the quasar that is likely accreting on to the galaxy group. We estimate inflow rates of 60–200 M⊙ yr−1, within an order of magnitude of the outflow rates in these systems. In the 4C 09.17 system, we detect narrow gas streams associated with satellite galaxies, potentially reminiscent of ram-pressure stripping seen in local galaxy groups and clusters. We find that the quasar host galaxies reside in dynamically complex environments, with ongoing mergers, gas accretion, ISM stripping, and outflows likely playing an important role in shaping the assembly and evolution of massive galaxies at cosmic noon.

Script-Independent Text Segmentation from Document Images

Document image analysis finds broad application in the digital world for the purpose of information retrieval. This includes optical character recognition (OCR), indexing of digital libraries, web image processing, etc. One of the important steps in this field is text segmentation. This segmentation becomes complicated for the documents containing text of uneven spacing and characters of varying font sizes. In this paper, script-independent text-line segmentation and word segmentation algorithms are presented. Fast marching method is used for text-line segmentation, whereas wavelet transform with connected components (CCs) labeling is used for word segmentation. Fast marching method is used as a region growing process that detects potential text-lines. For word segmentation, energy map is calculated using wavelet transform to create text-blocks. Both the proposed algorithms are evaluated on different databases containing documents of different scripts, where highest text-line and word segmentation accuracies of 98.9% and 99.1%, respectively, are obtained.

An improved algorithm for three-dimensional reconstruction of nonwoven microstructures by using multi-focus images

The filtration performance of nonwovens, such as virus protection and air purification, depends largely on their microstructure, that is, their internal structure. With the development of manufacturing techniques, nonwovens can be produced with complex microstructures to deliver low air resistance and high filtration efficiency. Therefore, it seems important to reconstruct an accurate three-dimensional (3D) model to describe the various microstructures for performance evaluation and product development. Multi-focus image fusion is one of the most effective ways used for the 3D nonwoven reconstruction. However, the overlapping and crossing of fibers in multi-focus images taken on an optional microscope make 3D fiber reconstruction incomplete and discontinuous. In this study, we developed a novel fiber segmentation and 3D nonwoven reconstruction algorithm to restore fiber continuity. The fiber web image was firstly thinned into the fiber mid-axis or skeleton, and fiber fragments were detected by searching the intersections. The features, such as the angle and depth, of each fiber fragment were obtained using Harris corner detection. A cluster technique was used to group the head-points of the fiber fragments that have similar features into one cluster. All fiber fragments in the same cluster are fitted with a polynomial curve to form a complete fiber in the 3D space. The experimental results show that this new method is suitable for the reconstruction of nonwovens with complex structures, such as fiber bonding or crossing, and can automatically detect and connect fiber fragments in 3D space. It provides an effective way for the subsequent simulation/evaluation of filtration characteristics.

Enhancement of automatic classification of arcus senilis-nonarcus senilis using convolutional neural networ

Cholesterol is a type of lipid found in the human body and is susceptible to abnormalities. It can be detected via lipid profiling through blood sampling. In addition, cholesterol can be detected through the presence of a "sodium ring" in the eye iris called the corneal arcus (CA), presenting a new preliminary detection method that is less invasive. Therefore, this paper proposed a non-invasive method in detecting cholesterol based on convolutional neural network (CNN) model representation using 300 normal and 300 abnormal iris images from UBIRIS and medical web images. In this work, contrast-limited adaptive histogram (CLAHE) and unsharp masking process was applied first on CA images to enhance the quality of CA images. To detect the CA images, the dataset was trained and tested using three pre-trained CNN architectures; one is created from scratch, another are Resnet-50 and VGG-19 architectures that were fine-tuned to the CA images. The best result was exhibited by proposed pre-trained CNN model created from scratch with 10-fold cross-validation that produced high average detection accuracy at 98.81%. Thus, deeper network implementation is recommended in the future to further improve CA localization for optometrists used in their daily clinical tasks in detecting cholesterol.

Research on the Filtering and Classification Method of Interactive Music Education Resources Based on Neural Network.

This work intends to classify and integrate music genres and emotions to improve the quality of music education. This work proposes a web image education resource retrieval method based on semantic network and interactive image filtering for a music education environment. It makes a judgment on these music source data and then uses these extracted feature sequences as the emotions expressed in the model of the combination of Long Short-Term Memory (LSTM) and Attention Mechanism (AM), thus judging the emotion category of music. The emotion recognition accuracy has increased after improving LSTM-AM into the BiGR-AM model. The greater the difference between emotion genres is, the easier it is to analyze the feature sequence containing emotion features, and the higher the recognition accuracy is. The classification accuracy of the excited, relieved, relaxed, and sad emotions can reach 76.5%, 71.3%, 80.8%, and 73.4%, respectively. The proposed interactive filtering method based on a Convolutional Recurrent Neural Network can effectively classify and integrate music resources to improve the quality of music education.

A LEARNING BASED CLASSIFICATION MODEL FOR VEGETABLES RECOGNITION AND RECIPES RECOMMENDATIONS

This paper presents the building of a classification model using Transfer learning, deployment of the model on the webpage, and recipes Recommendation using web scrapping. First, the vegetable classification model was built using transfer learning on the Pre-trained model MobilenetV2, which was originally trained on the Image Net dataset with 1.4 Million images and 1000 classes of web images. We used this as our base model to train our vegetable dataset and classify the images of 26 different vegetables. Second, the Model was deployed on a webpage using Tensor flow. JS. The model takes an image as input, classifies the image, and gives the vegetable name as output. Images can be taken from the webcam as well as from the storage. Third, based on model outputs, the web scraping script is used to fetch links for top-rated recipes from the internet.

A complex multiphase DLA associated with a compact group at z = 2.431 traces accretion, outflows, and tidal streams

ABSTRACT As part of our program to identify host galaxies of known z = 2–3 Mg ii absorbers with the Keck Cosmic Web Imager (KCWI), we discovered a compact group giving rise to a z = 2.431 DLA with ultrastrong Mg ii absorption in quasar field J234628+124859. The group consists of four star-forming galaxies within 8–28 kpc and v ∼ 40–340 km s−1 of each other, where tidal streams are weakly visible in deep HST imaging. The group geometric centre is D = 25 kpc from the quasar (D = 20–40 kpc for each galaxy). Galaxy G1 dominates the group (1.66L*, SFRFUV = 11.6 M⊙ yr−1) while G2, G3, and G4 are less massive (0.1–0.3L*, SFRFUV = 1.4–2.0 M⊙ yr−1). Using a VLT/UVES quasar spectrum covering the H i Lyman series and metal lines such as Mg ii, Si iii, and C iv, we characterized the kinematic structure and physical conditions along the line of sight with cloud-by-cloud multiphase Bayesian modelling. The absorption system has a total $\log (N({{{\rm H}\,\rm{\small I}}})/{\rm cm}^{-2})=20.53$ and an $N({{{\rm H}\,\rm{\small I}}})$-weighted mean metallicity of log (Z/Z⊙) = −0.68, with a very large Mg ii linewidth of Δv ∼ 700 km s−1. The highly kinematically complex profile is well modelled with 30 clouds across low- and intermediate-ionization phases with values ${13\lesssim \log (N({{{\rm H}\,\rm{\small I}}})/{\rm cm}^{-2})\lesssim 20}$ and −3 ≲ log (Z/Z⊙) ≲ 1. Comparing these properties to the galaxy properties, we infer a wide range of gaseous environments, including metal-rich outflows, metal-poor IGM accretion, and tidal streams from galaxy–galaxy interactions. This diversity of structures forms the intragroup medium around a complex compact group environment at the epoch of peak star formation activity. Surveys of low-redshift compact groups would benefit from obtaining a more complete census of this medium for characterizing evolutionary pathways.

Glo-In-One: holistic glomerular detection, segmentation, and lesion characterization with large-scale web image mining.

Purpose: The quantitative detection, segmentation, and characterization of glomeruli from high-resolution whole slide imaging (WSI) play essential roles in the computer-assisted diagnosis and scientific research in digital renal pathology. Historically, such comprehensive quantification requires extensive programming skills to be able to handle heterogeneous and customized computational tools. To bridge the gap of performing glomerular quantification for non-technical users, we develop the Glo-In-One toolkit to achieve holistic glomerular detection, segmentation, and characterization via a single line of command. Additionally, we release a large-scale collection of 30,000 unlabeled glomerular images to further facilitate the algorithmic development of self-supervised deep learning. Approach: The inputs of the Glo-In-One toolkit are WSIs, while the outputs are (1)WSI-level multi-class circle glomerular detection results (which can be directly manipulated with ImageScope), (2)glomerular image patches with segmentation masks, and (3)different lesion types. In the current version, the fine-grained global glomerulosclerosis (GGS) characterization is provided, including assessed-solidified-GSS (associated with hypertension-related injury), disappearing-GSS (a further end result of the SGGS becoming contiguous with fibrotic interstitium), and obsolescent-GSS (nonspecific GGS increasing with aging) glomeruli. To leverage the performance of the Glo-In-One toolkit, we introduce self-supervised deep learning to glomerular quantification via large-scale web image mining. Results: The GGS fine-grained classification model achieved a decent performance compared with baseline supervised methods while only using 10% of the annotated data. The glomerular detection achieved an average precision of 0.627 with circle representations, while the glomerular segmentation achieved a 0.955 patch-wise Dice dimilarity coefficient. Conclusion: We develop and release an open-source Glo-In-One toolkit, a software with holistic glomerular detection, segmentation, and lesion characterization. This toolkit is user-friendly to non-technical users via a single line of command. The toolbox and the 30,000 web mined glomerular images have been made publicly available at https://github.com/hrlblab/Glo-In-One.

Determination of multi-modal dimension metric learning with application to web image retrieval

Many real-world applications, like multimedia retrieval, confront the difficulty of determining the distance between any two items on Multi-modal data. According to most current Dimension Metrics Learning (DML) techniques, distance metrics may be learned using just one feature type or an aggregated feature space where many features are simply connected. Even though DML has been extensively researched. This study proposes a new framework for online learning, as well as a new classroom learning system for online Multi-modal dimension metric learning (OMDML), that is both efficient and scalable. This paper proposes a low-rank OMDML calculation to reduce the expensive cost of DML on high-dimensional component space, which reduces the computational cost while maintaining extremely competitive or much higher learning accuracy. With the purpose of determining whether or not multi-modular image recovery calculations can be successfully implemented, a large number of experiments are carried out. In most datasets tested, the suggested approach consistently outperforms alternative state-of-the-art algorithms, according to extensive experimental results

Multi-instance positive and unlabeled learning with bi-level embedding

Multiple Instance Learning (MIL) is a widely studied learning paradigm which arises from real applications. Existing MIL methods have achieved prominent performances under the premise of plenty annotation data. Nevertheless, sufficient labeled data is often unattainable due to the high labeling cost. For example, the task in web image identification is to find similar samples among a large size of unlabeled dataset through a small number of provided target pictures. This leads to a particular scenario of Multiple Instance Learning with insufficient Positive and superabundant Unlabeled data (PU-MIL), which is a hot research topic in MIL recently. In this paper, we propose a novel method called Multiple Instance Learning with Bi-level Embedding (MILBLE) to tackle PU-MIL problem. Unlike other PU-MIL method using only simple single-level mapping, the bi-level embedding strategy are designed to customize specific mapping for positive and unlabeled data. It ensures the characteristics of key instance are not erased. Moreover, the weighting measure adopted in positive data can extracts the uncontaminated information of true positive instances without interference from negative ones. Finally, we minimize the classification error loss of mapped examples based on class-prior probability to train the optimal classifier. Experimental results show that our method has better performance than other state-of-the-art methods.

Anime-to-real clothing: Cosplay costume generation via image-to-image translation

Cosplay has grown from its origins at fan conventions into a billion-dollar global dress phenomenon. To facilitate the imagination and reinterpretation of animated images as real garments, this paper presents an automatic costume-image generation method based on image-to-image translation. Cosplay items can be significantly diverse in their styles and shapes, and conventional methods cannot be directly applied to the wide variety of clothing images that are the focus of this study. To solve this problem, our method starts by collecting and preprocessing web images to prepare a cleaned, paired dataset of the anime and real domains. Then, we present a novel architecture for generative adversarial networks (GANs) to facilitate high-quality cosplay image generation. Our GAN consists of several effective techniques to bridge the two domains and improve both the global and local consistency of generated images. Experiments demonstrated that, with quantitative evaluation metrics, the proposed GAN performs better and produces more realistic images than conventional methods. Our codes and pretrained model are available on the web.

Kinematics and Feedback in H ii Regions in the Dwarf Starburst Galaxy IC 10

We present a survey of the central region of the nearest starburst galaxy, IC 10, using the W. M. Keck Observatory Keck Cosmic Web Imager (KCWI) at high spectral and spatial resolution. We map the central starburst of IC 10 to sample the kinematic and ionization properties of the individual star-forming regions. Using the low spectral resolution mode of KCWI, we map the oxygen abundance, and with the high spectral resolution mode, we identify 46 individual H II regions. These H II regions have an average radius of 4.0 pc, star formation rate ∼1.3 × 10−4 M ⊙ yr−1, and velocity dispersion ∼16 km s−1. None of the H II regions appear to be virialized (α vir ≫ 1), and on average, they show evidence of ongoing expansion. IC 10's H II regions are offset from the star-forming-region size–luminosity scaling relationships, as well as Larson’s Law that relates size and velocity dispersion. We investigate the balance of inward and outward pressure, P in and P out, finding P out > P in in 89% of H II regions, indicating feedback-driven expansion even in these low-mass H II regions. We find warm gas pressure (P gas) provides the dominant contribution to the outward pressure (P out). This counteracts the inward pressure, which is dominated by turbulence in the surrounding gas rather than self-gravity. Five H II regions show evidence of outflows that are most likely supported by either stellar winds (two regions) or champagne flows (three regions). These observations provide new insights into the state of the star-forming regions in IC 10 and negative feedback from low-mass clusters.

Robust Weakly Supervised Semantic Segmentation via Dual Domain Cascaded Decision and Annotation Bootstrap

The performance of weakly supervised semantic segmentation (WSSS) can be effectively improved by introducing Web images. In order to achieve knowledge transfer using external data robustly, a robust knowledge transfer learning method of dual domain collaborative bootstrap is proposed. Firstly, the knowledge transfer from Web domain to target domain is effectively achieved through dual domain cascaded decision-making strategy, which highly promotes the robustness of WSSS. In addition, the reliability is intensively enhanced due to reducing the noise of images and improving the quality of annotation by dual domain collaborative learning. The method achieves 65.4% and 65.9% mIoU on benchmark dataset PASCAL VOC 2012 validation and test sets, respectively. The experimental results outperforming the most of WSSS methods could show the effectiveness of the proposed method.

Towards Corruption-Agnostic Robust Domain Adaptation

Great progress has been achieved in domain adaptation in decades. Existing works are always based on an ideal assumption that testing target domains are independent and identically distributed with training target domains. However, due to unpredictable corruptions (e.g., noise and blur) in real data, such as web images and real-world object detection, domain adaptation methods are increasingly required to be corruption robust on target domains. We investigate a new task, corruption-agnostic robust domain adaptation (CRDA), to be accurate on original data and robust against unavailable-for-training corruptions on target domains. This task is non-trivial due to the large domain discrepancy and unsupervised target domains. We observe that simple combinations of popular methods of domain adaptation and corruption robustness have suboptimal CRDA results. We propose a new approach based on two technical insights into CRDA, as follows: (1) an easy-to-plug module called domain discrepancy generator (DDG) that generates samples that enlarge domain discrepancy to mimic unpredictable corruptions; (2) a simple but effective teacher-student scheme with contrastive loss to enhance the constraints on target domains. Experiments verify that DDG maintains or even improves its performance on original data and achieves better corruption robustness than baselines. Our code is available at: https://github.com/YifanXu74/CRDA .

The DUVET Survey: Resolved maps of star formation-driven outflows in a compact, starbursting disc galaxy

ABSTRACTWe study star formation-driven outflows in a z ∼ 0.02 starbursting disc galaxy, IRAS08339+6517, using spatially resolved measurements from the Keck Cosmic Web Imager (KCWI). We develop a new method incorporating a multistep process to determine whether an outflow should be fit in each spaxel, and then subsequently decompose the emission line into multiple components. We detect outflows ranging in velocity, vout, from 100 to 600 km s−1 across a range of star formation rate surface densities, ΣSFR, from ∼0.01 to 10 M⊙ yr−1 kpc−2 in resolution elements of a few hundred parsec. Outflows are detected in ∼100 per cent of all spaxels within the half-light radius, and ∼70 per cent within r90, suggestive of a high covering fraction for this starbursting disc galaxy. Around 2/3 of the total outflowing mass originates from the star forming ring, which corresponds to ${\lt}10{{\ \rm per\ cent}}$ of the total area of the galaxy. We find that the relationship between vout and the ΣSFR, as well as between the mass loading factor, η, and the ΣSFR, are consistent with trends expected from energy-driven feedback models. We study the resolution effects on this relationship and find stronger correlations above a re-binned size-scale of ∼500 pc. Conversely, we do not find statistically significant consistency with the prediction from momentum-driven winds.