检索结果-内蒙古大学图书馆

8th International Conference on Video and Image processing, ICVIP 2024

作者： Em, Tithnorakneath Aramvith, Supavadee Department of Electrical Engineering Faculty of Engineering Chulalongkorn University Bangkok Thailand Multimedia Data Analytics and Processing Research Unit Department of Electrical Engineering Faculty of Engineering Chulalongkorn University Bangkok Thailand

ISBN: (数字)9781510689244

ISBN: (纸本)9781510689237

Scene Text Image Super-Resolution (STISR) has become a significant pre-processing technique to enhance text recognition accuracy in low-resolution scene text images, particularly when addressing the challenges of low-resolution (LR) images in real-world applications. Despite advancements in general and text-focused methods, current techniques often fail to retain the fine-grained and text-specific details necessary for downstream recognition tasks. This paper presents a novel Attention-Guided Scene Text Image Super-Resolution (AG-STISR) framework designed to overcome these challenges. Our approach integrates advanced channel attention and spatial attention mechanisms within a shallow feature extraction module, enabling precise enhancement of key text features such as character edges, contours, and fine structural details. Channel attention helps prioritize the features rich in text-specific details by focusing the model's attention on the channels with the most important data for text recognition. On the other hand, Spatial attention emphasizes locating and highlighting the most significant spatial areas in the image, like character outlines and strokes, ensuring that key elements are maintained while minimizing background distractions. We extensively evaluated our model on the TextZoom dataset and assessed it against various scene text recognition standards. The results demonstrate that our AG-STISR model consistently outperforms state-of-the-art methods, showing significant improvements in PSNR, SSIM, and text recognition accuracy across varying levels of image degradation. This research introduces a compelling solution to STISR, enhancing not only image clarity but also the reliability of text recognition in real-world applications. © 2025 SPIE.

关键词： Image enhancement

来源：评论

学校读者我要写书评

暂无评论

2D T1-Weighted CE MRI Images Classification with Hybrid Channel Attention Block enhanced VGG-16 and MobileNet 1

2D T1-Weighted CE MRI Images Classification with Hybrid Chan...

引用

1st IEEE Karachi Section Humanitarian Technology Conference, Khi-HTC 2024

作者： Fareed, Mohammad Jawad Pluempitiwiriyawej, Charnchai Murtaza, Ghulam Chulalongkorn University Multimedia Data Analytics and Processing Research Unit Department of Electircal Engineering Bangkok Thailand

ISBN: (纸本)9798350373042

Motivated by the imperative need for precise and swift classification of brain tumors depicted in MRI images, this study unveils a novel classification network tailored explicitly for this task. Existing lightweight methodologies, although explored extensively, have fallen short of achieving the accuracy benchmarks attained by the model. The approach underwent rigorous testing on a challenging 2D T1-weighted CE-MRI dataset housing three distinct types of brain tumors: Meningioma, Glioma, and Pituitary. Recognizing the susceptibility of models to overfitting during training, introduced was a normalized combined channel and spatial attention mechanism, strategically incorporated to serve as an effective regularizer. Comparative analysis against state-of-the-art methodologies on this dataset highlights a notable performance boost of 2.12 percentage points achieved by integrating this attention mechanism into a base network. Furthermore, the fusion of the model with the pre-trained VGG16 network and GoogleNet into an ensemble demonstrates superior accuracy;however, this amalgamation comes at the expense of execution speed. © 2024 IEEE.

关键词： Image classification

来源：评论

学校读者我要写书评

暂无评论

Efficient Deep Attentive Pixels Network in Face Super-Resolution at Scale Factor of 16 20

Efficient Deep Attentive Pixels Network in Face Super-Resolu...

引用

20th International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology, ECTI-CON 2023

作者： Aung, Hein Htet Aramvith, Supavadee Faculty of Engineering Chulalongkorn University Multimedia Data Analytics and Processing Research Unit Department of Electrical Engineering Bangkok Thailand

ISBN: (纸本)9798350310467

Nowadays, Face Super-Resolution (FSR) models utilize the fusion approach, which combines the attention technique with the super-resolution network. The fusion approach has been proposed and solves the problem of FSR. Facial attributes have been effectively used to guide low-level information of the face to perform robust face image restoration. Iterative techniques appraised the value of facial landmarks to boost the reconstruction capability of the super-resolution network. Nevertheless, the network parameters in FSR are high, while the learning rate is still low. This paper proposes an attention mechanism combined with the Face Alignment Network (FAN). The proposed attention mechanism consists of channel attention and a non-local module. Our proposed model outperforms at the scale of $\times 16$ compared to the other state-of-the-art models. © 2023 IEEE.

关键词： Image reconstruction

来源：评论

学校读者我要写书评

暂无评论

A Survey of Orthographic Information in Machine Translation

引用

SN Computer Science 2021年第4期2卷 330页

作者： Chakravarthi, Bharathi Raja Rani, Priya Arcan, Mihael McCrae, John P. Unit for Linguistic Data Insight SFI Research Centre for Data Analytics Data Science Institute National University of Ireland Galway Galway Ireland Unit for Natural Language Processing Insight SFI Research Centre for Data Analytics Data Science Institute National University of Ireland Galway Galway Ireland

Machine translation is one of the applications of natural language processing which has been explored in different languages. Recently researchers started paying attention towards machine translation for resource-poor languages and closely related languages. A widespread and underlying problem for these machine translation systems is the linguistic difference and variation in orthographic conventions which causes many issues to traditional approaches. Two languages written in two different orthographies are not easily comparable but orthographic information can also be used to improve the machine translation system. This article offers a survey of research regarding orthography’s influence on machine translation of under-resourced languages. It introduces under-resourced languages in terms of machine translation and how orthographic information can be utilised to improve machine translation. We describe previous work in this area, discussing what underlying assumptions were made, and showing how orthographic knowledge improves the performance of machine translation of under-resourced languages. We discuss different types of machine translation and demonstrate a recent trend that seeks to link orthographic information with well-established machine translation methods. Considerable attention is given to current efforts using cognate information at different levels of machine translation and the lessons that can be drawn from this. Additionally, multilingual neural machine translation of closely related languages is given a particular focus in this survey. This article ends with a discussion of the way forward in machine translation with orthographic information, focusing on multilingual settings and bilingual lexicon induction. © 2021, The Author(s).

关键词： Machine translation Neural machine translation Orthography Rule-based machine translation Statistical machine translation Under-resourced languages

来源：评论

学校读者我要写书评

暂无评论

Non-Local Technique on Deep Attentive Face Super-Resolution Network 22

Non-Local Technique on Deep Attentive Face Super-Resolution ...

引用

22nd International Symposium on Communications and Information Technologies, ISCIT 2023

作者： Hajian, Amir Aung, Hein Htet Ruangsang, Watchara Chen, Sovann Aramvith, Supavadee Chulalongkorn University Department of Electrical Engineering Bangkok Thailand Multimedia Data Analytics and Processing Research Unit Department of Electrical Engineering Faculty of Engineering Chulalongkorn University Bangkok10330 Thailand

ISBN: (纸本)9781665457316

Recent Face Super-resolution (FSR) based on iterative collaboration between facial image recovery network and landmark estimation has succeeded in super-resolving facial images. However, the existing noise in coarse features at the low-level feature extraction leads to inaccurate facial priors such as landmarks and component maps, consequently degrading the super-resolved face image on a large scale. This paper proposes, a Non-local technique for deep attentive face super-resolution network (NLDA). A Non-local module has been designed before the residual channel attention block (RCAB) to eliminate noise degradation on coarse features effectively. The proposed model optimizes feature extraction and improves facial landmark fusion to yield higher-quality super-resolved images. This approach facilitates more accurate landmark estimation and boosts the performance of our model on a large scale and various face poses. Quantitative and qualitative experiments over CelebA and Helen face image datasets show that the proposed method outperforms other state-of-the-art FSR methods in recovering high-quality face images in various face poses and at a large scale. © 2023 IEEE.

关键词： Feature extraction

来源：评论

学校读者我要写书评

暂无评论

Deep Learning-Based Golf Swing Sequence Analysis 38

Deep Learning-Based Golf Swing Sequence Analysis

引用

38th IEEE Region 10 Conference, TENCON 2023

作者： Hajian, Amir Sookpreedee, Karit Phairoh, Kingrak Ruangsang, Watchara Aramvith, Supavadee Chulalongkorn University Department of Electrical Engineering Bangkok Thailand Chulalongkorn University Multimedia Data Analytics and Processing Research Unit Faculty of Engineering Department of Electrical Engineering Bangkok10330 Thailand

ISBN: (纸本)9798350302196

Golf is widely recognized as one of the most popular sports globally. However, one drawback of playing golf is the relatively high cost of equipment and coaching. While numerous training programs are available to assist players in their practice, there is currently no swing analysis program developed by Thai professionals. In this project, advanced deep learning models were employed: SwingNet, capable of predicting the sequence of eight golf swing events in videos and determining the confidence level of each swing, and MoveN et, designed to identify joint positions on the body and represent them as skeletons. These models were integrated into a customized template-matching algorithm that utilized angle-based measurements to analyze the sequence of golf swings. This analysis assessed the similarity score, represented as a percentage, between two individuals for each golf swing event. Furthermore, various techniques were implemented to enhance the efficiency of SwingN et. Through performance evaluation, it was observed that the efficiency of SwingN et surpassed by one percent compared to the pre-trained model. © 2023 IEEE.

关键词： Golf

来源：评论

学校读者我要写书评

暂无评论

Multiple Face Detection and Recognition for System-an-Chip FPGAs 21

Multiple Face Detection and Recognition for System-an-Chip F...

引用

21st International Joint Conference on Computer Science and Software Engineering, JCSSE 2024

作者： Santiwiwat, Siraphop Hajian, Amir Ruangsang, Watchara Aramvith, Supavadee Chulalongkorn University Multimedia Data Analytic and Processing Research Unit Faculty of Engineering Department of Electrical Engineering Bangkok Thailand

ISBN: (纸本)9798350381764

Recognizing multiple faces within a single frame or image presents a significant challenge in facial recognition tasks. This challenge demands robust algorithms capable of handling variations in facial position, unstable brightness, and the presence of accessories such as glasses or hats. However, in current surveillance applications, multiple face recognition is vital for maintaining various security measures. In such applications, there is a need for processors that are compact and deployable in diverse environments. FPGA is popular due to its compact size, cost-effectiveness, and high performance. This paper presents the implementation of multiple face detection and recognition on an SoC FPGA utilizing the Xilinx Kria KV260 platform. We chose the MTCNN algorithm for multiple face detection and the ResNet-20 model for multiple face recognition developed by Vitis AI, and we fine-tuned this model using the CUFace dataset. Our results demonstrate the robust capability to detect and recognize faces of up to 8 individuals within a single frame in just 0.11 seconds, with a recognition rate of 93.77%. © 2024 IEEE.

关键词： System-on-chip

来源：评论

学校读者我要写书评

暂无评论

Enhancing Visual Fidelity in Unpaired Night-to-Day Image Translation through a Perceptual Quality Focused Training Objective 7

Enhancing Visual Fidelity in Unpaired Night-to-Day Image Tra...

引用

7th SLAAI - International Conference on Artificial Intelligence, SLAAI-ICAI 2023

作者： Lakmal, H.K.I.S. Dissanayake, Maheshi B. Aramvith, Supavadee University of Peradeniya Faculty of Engineering Dept. of Electrical and Electronic Engineering Peradeniya Sri Lanka Chulalongkorn University Multimedia Data Analytics and Processing Research Unit Dept. of Electrical Engineering Bangkok Thailand

ISBN: (纸本)9798350319262

Nighttime driving poses visibility challenges, but image translation methods can help by transforming night images into day-like scenes. The Cycle-GAN is a versatile unpaired image translation model which can easily be adept to night-to-day image translation tasks. However, it generates unnatural and unrealistic outcomes in these cases. This study is focused on addressing this with a novel training strategy for the Cycle-GAN, employing a tailored training objective that incorporates perceptual quality optimization. This training objective aims to boost the naturalness and perceptual quality of the generated images by preserving high-level image features. The optimization process involves minimizing Euclidean distances between synthesized and target image feature maps, which are derived from the pre-trained VGG19 network. Experimental findings attest to the effectiveness of this method, revealing noteworthy improvements of 8% in Inception Score, 2% in NIQE Score, and 13% in BRISQUE Score. © 2023 IEEE.

关键词： Image enhancement

来源：评论

学校读者我要写书评

暂无评论

Compressed Image Super Resolution using Convolutional Neural Network 38

Compressed Image Super Resolution using Convolutional Neural...

引用

38th IEEE Region 10 Conference, TENCON 2023

作者： Tun, Ei Ei Konkitkriengkrai, Aktanin Ruangsang, Watchara Aramvith, Supavadee Chulalongkorn University Faculty of Engineering Department of Electrical Engineering Bangkok10330 Thailand Chulalongkorn University Multimedia Data Analytics and Processing Research Unit Faculty of Engineering Department of Electrical Engineering Bangkok10330 Thailand

ISBN: (纸本)9798350302196

Image compression is a topic of significant interest as it reduces file sizes in stored data. In this paper, we propose a model that achieves multiple levels of compression, thereby minimizing the storage space required for images, which typically consume substantial amounts of data due to their size and resolution. We combine an image downscaling and upscaling model with an image compression model. By leveraging convolutional techniques to identify image features, we can effectively reduce the size of the image through downscaling and subsequently upscaling it. Additionally, we employ entropy image compression and arithmetic encoding to compress and reconstruct the image while preserving its lossless data. Through experimentation with the Kodak dataset, we observed that our proposed model achieved a compression rate of 96.92%, significantly reducing the data needed for file storage. Moreover, our reconstructed images attained a standardized measure with a signal-to-noise ratio of 33.10 dB and a structural similarity of 0.9219. Notably, the perceptual quality of the images, including intricate details, remained intact to the human eye. © 2023 IEEE.

关键词： Image reconstruction

来源：评论

学校读者我要写书评

暂无评论

Combined Channel and Spatial Attention-Based Stereo Endoscopic Image Super-Resolution 38

Combined Channel and Spatial Attention-Based Stereo Endoscop...

引用

38th IEEE Region 10 Conference, TENCON 2023

作者： Hayat, Mansoor Armvith, Supavadee Achakulvisut, Titipat Chulalongkorn University Department of Electrical Engineering Bangkok Thailand Chulalongkorn University Multimedia Data Analytics and Processing Research Unit Department of Electrical Engineering Bangkok Thailand Mahidol University Department of Biomedical Engineering Bangkok Thailand

ISBN: (纸本)9798350302196

Stereo Imaging technology integration into medical diagnostics and surgeries brings a great revolution in the field of medical sciences. Now, surgeons and physicians have better insight into the anatomy of patients' organs. Like other technologies, stereo cameras have limitations, e.g., low resolution (LR) and blurry output images. Currently, most of the proposed techniques for super-resolution focus on developing complex blocks and complicated loss functions, which cause high system complexity. We proposed a combined channel and spatial attention block to extract features incorporated with a specific but very strong parallax attention module (PAM) for endoscopic image super-resolution. The proposed model is trained using the da Vinci dataset on scales 2 and 4. Our proposed model has improved PSNR up to 2.12 dB for scale 2 and 1.29 dB for scale 4, while SSIM is improved by 0.03 for scale 2 and 0.0008 for scale 4. By incorporating this method, diagnosis and treatment for endoscopic images can be more accurate and effective. © 2023 IEEE.

关键词： Geometrical optics

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：