Pub Date : 2021-08-19DOI: 10.1109/RIVF51545.2021.9642138
Bao Ngoc Vi, Dinh Tan Nguyen, Cao Truong Tran, Huu Phuc Ngo, Chi Cong Nguyen, Hai-Hong Phan
Missing values present as the most common problem in real-world data science. Inadequate treatment of missing values could often result in mass errors. Hence missing values should be managed conscientiously for classification. Generative Adversarial Networks (GANs) have been applied for imputing missing values in most recent years. This paper proposes a multiple imputation method to estimate missing values for classification through the integration of GAN and ensemble learning. Our propose method MIGAN utilises GAN to generate different training observations which are then used to conduct ensemble classifiers for classification with missing data. We conducted our experiments examine MIGAN on various data sets as well as comparing MIGAN with the state-of-the-art imputation methods. The experimental results show significant results, which highlights the accuracy of MIGAN in classifying the missing data.
{"title":"Multiple Imputation by Generative Adversarial Networks for Classification with Incomplete Data","authors":"Bao Ngoc Vi, Dinh Tan Nguyen, Cao Truong Tran, Huu Phuc Ngo, Chi Cong Nguyen, Hai-Hong Phan","doi":"10.1109/RIVF51545.2021.9642138","DOIUrl":"https://doi.org/10.1109/RIVF51545.2021.9642138","url":null,"abstract":"Missing values present as the most common problem in real-world data science. Inadequate treatment of missing values could often result in mass errors. Hence missing values should be managed conscientiously for classification. Generative Adversarial Networks (GANs) have been applied for imputing missing values in most recent years. This paper proposes a multiple imputation method to estimate missing values for classification through the integration of GAN and ensemble learning. Our propose method MIGAN utilises GAN to generate different training observations which are then used to conduct ensemble classifiers for classification with missing data. We conducted our experiments examine MIGAN on various data sets as well as comparing MIGAN with the state-of-the-art imputation methods. The experimental results show significant results, which highlights the accuracy of MIGAN in classifying the missing data.","PeriodicalId":6860,"journal":{"name":"2021 RIVF International Conference on Computing and Communication Technologies (RIVF)","volume":"15 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2021-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"73705059","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-08-19DOI: 10.1109/RIVF51545.2021.9642150
C. M. Nguyen, Vi Van Ngo, Dang Duy Nguyen
This challenge organized at the RIVF conference 2021 [12], with two tasks including (1) image quality assessment (IQA) of the captured receipt, and (2) key information extraction (KIE) of required fields, our team came up with a solution based on extracting image patches for task 1 and Yolov5 + VietOCR for task 2. Our solution achieved 0.149 of the RMSE score for task 1 (rank 7) and 0.219 of the CER score for task 2 (rank 1). Our code is available at https://github.com/cuongngm/RIVF2021.
{"title":"MC-OCR Challenge 2021: Simple approach for receipt information extraction and quality evaluation","authors":"C. M. Nguyen, Vi Van Ngo, Dang Duy Nguyen","doi":"10.1109/RIVF51545.2021.9642150","DOIUrl":"https://doi.org/10.1109/RIVF51545.2021.9642150","url":null,"abstract":"This challenge organized at the RIVF conference 2021 [12], with two tasks including (1) image quality assessment (IQA) of the captured receipt, and (2) key information extraction (KIE) of required fields, our team came up with a solution based on extracting image patches for task 1 and Yolov5 + VietOCR for task 2. Our solution achieved 0.149 of the RMSE score for task 1 (rank 7) and 0.219 of the CER score for task 2 (rank 1). Our code is available at https://github.com/cuongngm/RIVF2021.","PeriodicalId":6860,"journal":{"name":"2021 RIVF International Conference on Computing and Communication Technologies (RIVF)","volume":"86 1","pages":"1-4"},"PeriodicalIF":0.0,"publicationDate":"2021-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74199475","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-08-19DOI: 10.1109/RIVF51545.2021.9642066
Vu Hiep Dao, Hieu Mac, Duc Tran
Low light conditions are known to create a notable challenge to the applicability of deep learning in a wide variety of computer vision applications. In this paper, we develop a detection method for real-time multispectral pedestrians that fuses color image (i.e., red-green-blue or RBG) with thermal image to provide a reliable object vision. Such combination is achieved using the confidence scores that are computed based on the illumination measure of a given input image. We evaluate the proposed algorithm on KAIST dataset. Such method is observed to give a 34.11% Log Average Miss Rate, operate in real-time, and thus, being ready to deploy in practice.
{"title":"A Real-time Multispectral Algorithm for Robust Pedestrian Detection","authors":"Vu Hiep Dao, Hieu Mac, Duc Tran","doi":"10.1109/RIVF51545.2021.9642066","DOIUrl":"https://doi.org/10.1109/RIVF51545.2021.9642066","url":null,"abstract":"Low light conditions are known to create a notable challenge to the applicability of deep learning in a wide variety of computer vision applications. In this paper, we develop a detection method for real-time multispectral pedestrians that fuses color image (i.e., red-green-blue or RBG) with thermal image to provide a reliable object vision. Such combination is achieved using the confidence scores that are computed based on the illumination measure of a given input image. We evaluate the proposed algorithm on KAIST dataset. Such method is observed to give a 34.11% Log Average Miss Rate, operate in real-time, and thus, being ready to deploy in practice.","PeriodicalId":6860,"journal":{"name":"2021 RIVF International Conference on Computing and Communication Technologies (RIVF)","volume":"18 4 1","pages":"1-4"},"PeriodicalIF":0.0,"publicationDate":"2021-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"83455735","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-08-19DOI: 10.1109/RIVF51545.2021.9642098
Bao Ngoc Vi, Nguyen Ngoc Tran, Trung Giap Vu The
Password is the most widely used authenticate method. Individuals ordinarily have numerous passwords for their documents or devices, and, in some cases, they need to recover them with password guessing tools. Most popular guessing tools require a dictionary of common passwords to check with password hashes. Thus, generative adversarial networks (GANs) are suitable choices to automatically create a high-quality dictionary without any additional information from experts or password structures. One of the successful GAN-based models is the PassGAN. However, existing GAN-based models suffer from the discrete nature of passwords. Therefore, we proposed and evaluated two improvement of the PassGAN model to tackle this problem: the GS-PassGAN model using Gumbel-Softmax relaxation and the S-PassGAN using a smooth representation of a real password obtained by an additional Auto-Encoder. Experiment results on three different popular datasets show that the proposed method is better than the PassGAN both in the standalone and combining cases. Moreover, the matching rate of the proposed method can be increased by more than 5%.
{"title":"A GAN-based approach for password guessing","authors":"Bao Ngoc Vi, Nguyen Ngoc Tran, Trung Giap Vu The","doi":"10.1109/RIVF51545.2021.9642098","DOIUrl":"https://doi.org/10.1109/RIVF51545.2021.9642098","url":null,"abstract":"Password is the most widely used authenticate method. Individuals ordinarily have numerous passwords for their documents or devices, and, in some cases, they need to recover them with password guessing tools. Most popular guessing tools require a dictionary of common passwords to check with password hashes. Thus, generative adversarial networks (GANs) are suitable choices to automatically create a high-quality dictionary without any additional information from experts or password structures. One of the successful GAN-based models is the PassGAN. However, existing GAN-based models suffer from the discrete nature of passwords. Therefore, we proposed and evaluated two improvement of the PassGAN model to tackle this problem: the GS-PassGAN model using Gumbel-Softmax relaxation and the S-PassGAN using a smooth representation of a real password obtained by an additional Auto-Encoder. Experiment results on three different popular datasets show that the proposed method is better than the PassGAN both in the standalone and combining cases. Moreover, the matching rate of the proposed method can be increased by more than 5%.","PeriodicalId":6860,"journal":{"name":"2021 RIVF International Conference on Computing and Communication Technologies (RIVF)","volume":"16 1","pages":"1-5"},"PeriodicalIF":0.0,"publicationDate":"2021-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90542453","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-08-19DOI: 10.1109/RIVF51545.2021.9642102
D. Nguyen, Do Minh Kha, Pham Thi To Nga, Pham Ngoc Hung
This paper presents an autoencoder-based method for a targeted attack on deep neural network models, named AE4DNN. The proposed method aims to improve the existing targeted attacks in terms of their generalization, transferability, and the trade-off between the quality of adversarial examples and the computational cost. The idea of AE4DNN is that an autoencoder model is trained from a balanced subset of the training set. The trained autoencoder model is then used to generate adversarial examples from the remaining subset of the training set, produce adversarial examples from new samples, and attack other DNN models. To demonstrate the effectiveness of AE4DNN, the compared methods are box-constrained L-BFGS, Carlini-Wagner ‖L‖2 attack, and AAE. The comprehensive experiment on MNIST has shown that AE4DNN can gain a better transferability, improve generalization, and generate high quality of adversarial examples while requiring a low cost of computation. This initial result demonstrates the potential ability of AE4DNN in practice, which would help to reduce the effort of testing deep neural network models.
{"title":"An Autoencoder-based Method for Targeted Attack on Deep Neural Network Models","authors":"D. Nguyen, Do Minh Kha, Pham Thi To Nga, Pham Ngoc Hung","doi":"10.1109/RIVF51545.2021.9642102","DOIUrl":"https://doi.org/10.1109/RIVF51545.2021.9642102","url":null,"abstract":"This paper presents an autoencoder-based method for a targeted attack on deep neural network models, named AE4DNN. The proposed method aims to improve the existing targeted attacks in terms of their generalization, transferability, and the trade-off between the quality of adversarial examples and the computational cost. The idea of AE4DNN is that an autoencoder model is trained from a balanced subset of the training set. The trained autoencoder model is then used to generate adversarial examples from the remaining subset of the training set, produce adversarial examples from new samples, and attack other DNN models. To demonstrate the effectiveness of AE4DNN, the compared methods are box-constrained L-BFGS, Carlini-Wagner ‖L‖2 attack, and AAE. The comprehensive experiment on MNIST has shown that AE4DNN can gain a better transferability, improve generalization, and generate high quality of adversarial examples while requiring a low cost of computation. This initial result demonstrates the potential ability of AE4DNN in practice, which would help to reduce the effort of testing deep neural network models.","PeriodicalId":6860,"journal":{"name":"2021 RIVF International Conference on Computing and Communication Technologies (RIVF)","volume":"8 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2021-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89185825","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-08-19DOI: 10.1109/RIVF51545.2021.9642067
Nguyễn Anh Cường, D. Mai, Do Viet Duc, Trong Hop Dang, L. Ngo, L. T. Pham
For clustering problems, each data sample has the potential to belong to many different clusters depending on the similarity. However, besides the degree of similarity and non-similarity, there is a degree of hesitation in determining whether or not a data sample belongs to a defined cluster. Besides the fuzzy c-means algorithm (FCM), another popular algorithm is fuzzy C-medoids clustering (FCMdd). FCMdd chooses several existing objects as the cluster centroids, while FCM considers the samples’ weighted average to be the cluster centroid. This subtle difference causes the FCMdd is more resistant to interference than FCM. Since noise samples will more easily affect the center of centroids of the FCM, it is easier to create clustering results with great accuracy. In this study, we proposed a method for extending the fuzzy c-medoids clustering based on interval type-2 intuitionistic fuzzy sets, named the interval type-2 intuitionistic fuzzy c-medoids clustering algorithm (IT2IFCMdd). With this combination, the proposed algorithm can take advantage of both the fuzzy c-medoids clustering (FCMdd) method and the interval type-2 intuitionistic fuzzy sets applied to the clustering problem. Experiments performed on data sets commonly used in machine learning show that the proposed method gives better clustering results in most experimental cases.
{"title":"Fuzzy C-Medoids Clustering Based on Interval Type-2 Inituitionistic Fuzzy Sets","authors":"Nguyễn Anh Cường, D. Mai, Do Viet Duc, Trong Hop Dang, L. Ngo, L. T. Pham","doi":"10.1109/RIVF51545.2021.9642067","DOIUrl":"https://doi.org/10.1109/RIVF51545.2021.9642067","url":null,"abstract":"For clustering problems, each data sample has the potential to belong to many different clusters depending on the similarity. However, besides the degree of similarity and non-similarity, there is a degree of hesitation in determining whether or not a data sample belongs to a defined cluster. Besides the fuzzy c-means algorithm (FCM), another popular algorithm is fuzzy C-medoids clustering (FCMdd). FCMdd chooses several existing objects as the cluster centroids, while FCM considers the samples’ weighted average to be the cluster centroid. This subtle difference causes the FCMdd is more resistant to interference than FCM. Since noise samples will more easily affect the center of centroids of the FCM, it is easier to create clustering results with great accuracy. In this study, we proposed a method for extending the fuzzy c-medoids clustering based on interval type-2 intuitionistic fuzzy sets, named the interval type-2 intuitionistic fuzzy c-medoids clustering algorithm (IT2IFCMdd). With this combination, the proposed algorithm can take advantage of both the fuzzy c-medoids clustering (FCMdd) method and the interval type-2 intuitionistic fuzzy sets applied to the clustering problem. Experiments performed on data sets commonly used in machine learning show that the proposed method gives better clustering results in most experimental cases.","PeriodicalId":6860,"journal":{"name":"2021 RIVF International Conference on Computing and Communication Technologies (RIVF)","volume":"11 1","pages":"1-5"},"PeriodicalIF":0.0,"publicationDate":"2021-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89561300","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-08-19DOI: 10.1109/RIVF51545.2021.9642070
Nguyen Ba Hung, Thanh Duc Nguyen, Thai Van Chien, D. V. Sang
Colorectal cancer is one of the most prevalent causes of cancer-related death. Early polyp segmentation in colonoscopy is helpful in diagnosing and preventing colorectal cancer. However, this task a challenging due to variations in the appearance of polyps. This paper proposes a new encoder-decoder-based method called AG-ResUNet++ that leverages attention gate mechanism and residual connections to enhance the performance of the existing UNet++ model in the polyp segmentation task. Our method considerably outperforms other state-of-the-art methods on the popular polyp segmentation datasets, including KvasirSEG and CVC-612.
{"title":"AG-ResUNet++: An Improved Encoder-Decoder Based Method for Polyp Segmentation in Colonoscopy Images","authors":"Nguyen Ba Hung, Thanh Duc Nguyen, Thai Van Chien, D. V. Sang","doi":"10.1109/RIVF51545.2021.9642070","DOIUrl":"https://doi.org/10.1109/RIVF51545.2021.9642070","url":null,"abstract":"Colorectal cancer is one of the most prevalent causes of cancer-related death. Early polyp segmentation in colonoscopy is helpful in diagnosing and preventing colorectal cancer. However, this task a challenging due to variations in the appearance of polyps. This paper proposes a new encoder-decoder-based method called AG-ResUNet++ that leverages attention gate mechanism and residual connections to enhance the performance of the existing UNet++ model in the polyp segmentation task. Our method considerably outperforms other state-of-the-art methods on the popular polyp segmentation datasets, including KvasirSEG and CVC-612.","PeriodicalId":6860,"journal":{"name":"2021 RIVF International Conference on Computing and Communication Technologies (RIVF)","volume":"14 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2021-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"82405972","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-08-19DOI: 10.1109/RIVF51545.2021.9642077
Xuan-Son Vu, Quang-Anh Bui, Nhu-Van Nguyen, Thi-Tuyet-Hai Nguyen, Thanh Vu
The paper describes the organisation of the "Mobile Captured Receipt Recognition Challenge" (MC-OCR) task at the RIVF conference 2021 1 on recognizing the fine-grained information in Vietnamese receipts captured using mobile devices. The task is organized as a multi-tasking model on a dataset containing 2,436 Vietnamese receipts. The participants were challenged to build a model that is capable of (1) predicting receipt’s quality based on readable information, and (2) recognizing textual information of four required information (i.e., "SELLER", "SELLER ADDRESS", "TIMESTAMP", and "TOTAL COST") in the receipts. MC-OCR challenge happened in one month and top winners of each task will present their solutions at RIVF 2021. Participants were competing on CodaLab.Org from 05th December 2020 to 23rd January 2021. All participants with valid submitted results were encouraged to submit their papers. Within one month, the challenge has attracted 105 participants and recorded about 1,285 submission entries.
{"title":"MC-OCR Challenge: Mobile-Captured Image Document Recognition for Vietnamese Receipts","authors":"Xuan-Son Vu, Quang-Anh Bui, Nhu-Van Nguyen, Thi-Tuyet-Hai Nguyen, Thanh Vu","doi":"10.1109/RIVF51545.2021.9642077","DOIUrl":"https://doi.org/10.1109/RIVF51545.2021.9642077","url":null,"abstract":"The paper describes the organisation of the \"Mobile Captured Receipt Recognition Challenge\" (MC-OCR) task at the RIVF conference 2021 1 on recognizing the fine-grained information in Vietnamese receipts captured using mobile devices. The task is organized as a multi-tasking model on a dataset containing 2,436 Vietnamese receipts. The participants were challenged to build a model that is capable of (1) predicting receipt’s quality based on readable information, and (2) recognizing textual information of four required information (i.e., \"SELLER\", \"SELLER ADDRESS\", \"TIMESTAMP\", and \"TOTAL COST\") in the receipts. MC-OCR challenge happened in one month and top winners of each task will present their solutions at RIVF 2021. Participants were competing on CodaLab.Org from 05th December 2020 to 23rd January 2021. All participants with valid submitted results were encouraged to submit their papers. Within one month, the challenge has attracted 105 participants and recorded about 1,285 submission entries.","PeriodicalId":6860,"journal":{"name":"2021 RIVF International Conference on Computing and Communication Technologies (RIVF)","volume":"1 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2021-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"79144459","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-08-19DOI: 10.1109/RIVF51545.2021.9642079
Hai Cao, Viet-Trung Tran
Address standardization is the process of converting and mapping free-form addresses into a standard structured format. For many business cases, the addresses are entered into the information systems by end-users. They are often noisy, uncompleted, and in different formatted styles. In this paper, we propose a deep learning-based approach to the address standardization challenge. Our key idea is to leverage a Siamese neural network model to embed raw inputs and standardized addresses into a single latent multi-dimensional space. Thus, the corresponding of the raw input address is the one with the highest-ranking score. Our experiments demonstrate that our best model achieved 95.41% accuracy, which is 6.6% improvement from the current state of the art.
{"title":"Deep neural network based learning to rank for address standardization","authors":"Hai Cao, Viet-Trung Tran","doi":"10.1109/RIVF51545.2021.9642079","DOIUrl":"https://doi.org/10.1109/RIVF51545.2021.9642079","url":null,"abstract":"Address standardization is the process of converting and mapping free-form addresses into a standard structured format. For many business cases, the addresses are entered into the information systems by end-users. They are often noisy, uncompleted, and in different formatted styles. In this paper, we propose a deep learning-based approach to the address standardization challenge. Our key idea is to leverage a Siamese neural network model to embed raw inputs and standardized addresses into a single latent multi-dimensional space. Thus, the corresponding of the raw input address is the one with the highest-ranking score. Our experiments demonstrate that our best model achieved 95.41% accuracy, which is 6.6% improvement from the current state of the art.","PeriodicalId":6860,"journal":{"name":"2021 RIVF International Conference on Computing and Communication Technologies (RIVF)","volume":"11 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2021-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84177520","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-08-19DOI: 10.1109/RIVF51545.2021.9642145
Anh Tuan Vuong, Van Ha Tang, L. Ngo
The hyperspectral image (HSI) can provide useful information about the desired objects using spectral, spatial, and band channels. However, the image quality is typically distorted due to the limitations of sensing conditions and hardware operations. Consequently, the HSI is typically contaminated by a mixture noise during the acquisition process, including dead lines, stripes, Gaussian noise and impulse noise. In this paper, we introduce a new denoising model based on low-rank matrix recovery (LRMR), which can effectively remove various kinds of noise in HSI data. The low-rank property of the hyperspectral imagery is exploited by converting a patch of the HSI data from 3-D matrix into a 2-D matrix. The dead lines, stripes, and impulse noise are all modelled as sparse noise. To efficiently remove mixed noise and enhance performance, we develop an iterative algorithm using greedy bilateral technique to solve the optimization problem. To illustrate the proposed method’s efficacy in restoring HSI, both simulated and real-world HSI experiments are conducted.
{"title":"A Hyperspectral Image Denoising Approach via Low-Rank Matrix Recovery and Greedy Bilateral","authors":"Anh Tuan Vuong, Van Ha Tang, L. Ngo","doi":"10.1109/RIVF51545.2021.9642145","DOIUrl":"https://doi.org/10.1109/RIVF51545.2021.9642145","url":null,"abstract":"The hyperspectral image (HSI) can provide useful information about the desired objects using spectral, spatial, and band channels. However, the image quality is typically distorted due to the limitations of sensing conditions and hardware operations. Consequently, the HSI is typically contaminated by a mixture noise during the acquisition process, including dead lines, stripes, Gaussian noise and impulse noise. In this paper, we introduce a new denoising model based on low-rank matrix recovery (LRMR), which can effectively remove various kinds of noise in HSI data. The low-rank property of the hyperspectral imagery is exploited by converting a patch of the HSI data from 3-D matrix into a 2-D matrix. The dead lines, stripes, and impulse noise are all modelled as sparse noise. To efficiently remove mixed noise and enhance performance, we develop an iterative algorithm using greedy bilateral technique to solve the optimization problem. To illustrate the proposed method’s efficacy in restoring HSI, both simulated and real-world HSI experiments are conducted.","PeriodicalId":6860,"journal":{"name":"2021 RIVF International Conference on Computing and Communication Technologies (RIVF)","volume":"31 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2021-08-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"88457171","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：