Pub Date : 2019-06-01DOI: 10.1109/icce-asia46551.2019.8942212
Yuzana Win
As a growth of the technological world, web technologies and social networking emerged and played an important role in telecommunication. People misuse the social network as a new weapon to make a person attack unable to find the identity of the attacker. Due to the illegal action, the technological world seems to face new challenges and new risks like cyberbullying. This paper proposes a supervised method for detection of Myanmar cyberbullying in social media by using Support Vector Machine (SVM) classifier. The proposed method includes three main steps: data preprocessing, word segmentation, and classification. In the first step, we extract the posts written in Myanmar text from social media. We break the posts and sentences into syllables into words by using the Longest Syllable Matching approach along with a dictionary as the second step. For the third step, we apply Support Vector Machine classifier to detect cyberbullying in social media whether the bullying words or not. Consequently, the experimental result shows that our method obtains 0.7540 classification accuracy in terms of F-score.
{"title":"Classification using Support Vector Machine to Detect Cyberbullying in Social Media for Myanmar Language","authors":"Yuzana Win","doi":"10.1109/icce-asia46551.2019.8942212","DOIUrl":"https://doi.org/10.1109/icce-asia46551.2019.8942212","url":null,"abstract":"As a growth of the technological world, web technologies and social networking emerged and played an important role in telecommunication. People misuse the social network as a new weapon to make a person attack unable to find the identity of the attacker. Due to the illegal action, the technological world seems to face new challenges and new risks like cyberbullying. This paper proposes a supervised method for detection of Myanmar cyberbullying in social media by using Support Vector Machine (SVM) classifier. The proposed method includes three main steps: data preprocessing, word segmentation, and classification. In the first step, we extract the posts written in Myanmar text from social media. We break the posts and sentences into syllables into words by using the Longest Syllable Matching approach along with a dictionary as the second step. For the third step, we apply Support Vector Machine classifier to detect cyberbullying in social media whether the bullying words or not. Consequently, the experimental result shows that our method obtains 0.7540 classification accuracy in terms of F-score.","PeriodicalId":117814,"journal":{"name":"2019 IEEE International Conference on Consumer Electronics - Asia (ICCE-Asia)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121128052","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 : 2019-06-01DOI: 10.1109/icce-asia46551.2019.8942202
Yeongeun Seo, Jaehyun Park, Jung Ho Ahn, Taesup Moon
Branch predictor is a critical component in CPUs because its prediction accuracy highly influences the performance of computer devices. This technology attempts to predict whether a branch instruction is ‘taken’ or ‘not taken’ and executes the following instructions in an execution order based on the prediction result. If the prediction is incorrect, those speculatively executed instructions must be rolled back, causing overheads on both performance and energy efficiency. Conventional branch predictors typically adopt rule-based methods exploiting branch history (i.e., whether recently encountered branches in the course of execution or on the same address of the current instruction were taken or not), whereas deep learning-based prediction methods have been recently proposed. In this paper, we show the neural network model learned with less dataset generalizes well for all applications, not just for specific applications in the training set. Also, unlike the previous deep learning-based branch prediction studies, which were difficult to reproduce, this paper includes clear experiment contents.
{"title":"Exploring Deep Learning-based Branch Prediction for Computer Devices","authors":"Yeongeun Seo, Jaehyun Park, Jung Ho Ahn, Taesup Moon","doi":"10.1109/icce-asia46551.2019.8942202","DOIUrl":"https://doi.org/10.1109/icce-asia46551.2019.8942202","url":null,"abstract":"Branch predictor is a critical component in CPUs because its prediction accuracy highly influences the performance of computer devices. This technology attempts to predict whether a branch instruction is ‘taken’ or ‘not taken’ and executes the following instructions in an execution order based on the prediction result. If the prediction is incorrect, those speculatively executed instructions must be rolled back, causing overheads on both performance and energy efficiency. Conventional branch predictors typically adopt rule-based methods exploiting branch history (i.e., whether recently encountered branches in the course of execution or on the same address of the current instruction were taken or not), whereas deep learning-based prediction methods have been recently proposed. In this paper, we show the neural network model learned with less dataset generalizes well for all applications, not just for specific applications in the training set. Also, unlike the previous deep learning-based branch prediction studies, which were difficult to reproduce, this paper includes clear experiment contents.","PeriodicalId":117814,"journal":{"name":"2019 IEEE International Conference on Consumer Electronics - Asia (ICCE-Asia)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130878500","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 : 2019-06-01DOI: 10.1109/icce-asia46551.2019.8941604
Jennifer C. Dela Cruz, Ramon G. Garcia, Annissa Vi C. Diaz, Angelika Mae B. Diño, Danielle Jane I. Nicdao, Christine Shayne S. Venancio
Blood type can be determined by the presence or absence of antigens in the red blood cells, and can be classified by the ABO (A, $B$, AB, O) and Rh D (either positive or negative) systems. Knowing one's blood type is one of the most crucial steps before blood transfusion or any medical operations to prevent the risk of receiving incompatible blood that could lead to adverse or even fatal reactions to patients. Although fully automated blood testing instruments are already being used in some major hospitals, its large size and long processing time, limit its ability to be used in emergency situations. Hence, during onsite blood typing, the traditional or the slide method is being used, which is less accurate due to human errors. This paper presents a raspberry pi based image processing system that is capable of determining all eight types of blood using Canny Edge and Contour Detection. All blood types detected by the proposed system matched that of the known blood samples for the controlled testing of all five samples with five trials each sample for the known A+, $B$+ AB+, O+, A-, B-, AB- and O-. Uncontrolled testing was also performed to compare the results of the ten random blood types identified by the proposed prototype to the results obtained from test tube method. All these ten samples matched the results obtained from the clinical laboratory. This portable and automated device could avoid human errors, without risking accurate results that could be obtain in a short period of time.
{"title":"Portable Blood Typing Device Using Image Analysis","authors":"Jennifer C. Dela Cruz, Ramon G. Garcia, Annissa Vi C. Diaz, Angelika Mae B. Diño, Danielle Jane I. Nicdao, Christine Shayne S. Venancio","doi":"10.1109/icce-asia46551.2019.8941604","DOIUrl":"https://doi.org/10.1109/icce-asia46551.2019.8941604","url":null,"abstract":"Blood type can be determined by the presence or absence of antigens in the red blood cells, and can be classified by the ABO (A, $B$, AB, O) and Rh D (either positive or negative) systems. Knowing one's blood type is one of the most crucial steps before blood transfusion or any medical operations to prevent the risk of receiving incompatible blood that could lead to adverse or even fatal reactions to patients. Although fully automated blood testing instruments are already being used in some major hospitals, its large size and long processing time, limit its ability to be used in emergency situations. Hence, during onsite blood typing, the traditional or the slide method is being used, which is less accurate due to human errors. This paper presents a raspberry pi based image processing system that is capable of determining all eight types of blood using Canny Edge and Contour Detection. All blood types detected by the proposed system matched that of the known blood samples for the controlled testing of all five samples with five trials each sample for the known A+, $B$+ AB+, O+, A-, B-, AB- and O-. Uncontrolled testing was also performed to compare the results of the ten random blood types identified by the proposed prototype to the results obtained from test tube method. All these ten samples matched the results obtained from the clinical laboratory. This portable and automated device could avoid human errors, without risking accurate results that could be obtain in a short period of time.","PeriodicalId":117814,"journal":{"name":"2019 IEEE International Conference on Consumer Electronics - Asia (ICCE-Asia)","volume":"38 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114726619","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 : 2019-06-01DOI: 10.1109/icce-asia46551.2019.8942201
A. M. Tun, May Aye Khine
Tropical cyclones threatened many countries around the Bay of Bengal as storm surges. India, Bangladesh, and Myanmar have much destruction along the coastal regions due to storm surge. So, storm surge prediction needs to be accurate. Traditional process-based numerical models have high computational demands to make timely forecast and deterministic numerical models are strongly dependent on accurate meteorological input to predict storm surge. In this work, a Long Short-Term Memory Neural Network (LSTM) used to simulate the maximum sustained wind speed of storm in coastal areas of the Bay of Bengal and the Arabian Sea. Simulated and historical storm data are collected from the Regional Specialized Meteorological Centre (RSMC).
{"title":"Maximum Sustained Wind Speed Simulation of Storm Surge with Long Short-Term Memory","authors":"A. M. Tun, May Aye Khine","doi":"10.1109/icce-asia46551.2019.8942201","DOIUrl":"https://doi.org/10.1109/icce-asia46551.2019.8942201","url":null,"abstract":"Tropical cyclones threatened many countries around the Bay of Bengal as storm surges. India, Bangladesh, and Myanmar have much destruction along the coastal regions due to storm surge. So, storm surge prediction needs to be accurate. Traditional process-based numerical models have high computational demands to make timely forecast and deterministic numerical models are strongly dependent on accurate meteorological input to predict storm surge. In this work, a Long Short-Term Memory Neural Network (LSTM) used to simulate the maximum sustained wind speed of storm in coastal areas of the Bay of Bengal and the Arabian Sea. Simulated and historical storm data are collected from the Regional Specialized Meteorological Centre (RSMC).","PeriodicalId":117814,"journal":{"name":"2019 IEEE International Conference on Consumer Electronics - Asia (ICCE-Asia)","volume":"87 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126729060","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 : 2019-06-01DOI: 10.1109/icce-asia46551.2019.8942196
Suk-Un Yoon, Jinho Kim, Jongha Woo, Younghoon Moon, Cheul-hee Hahm
In this paper, we introduce a novel service, Ambient Mode, which delivers a functional value to customers when TV is off. Instead of showing a meaningless black screen, the Ambient Mode provides meaningful and emotional background experiences. To mitigate users' concern about the cost of using the Ambient feature, the TV intelligently detects the presence of a person in the room from the registered mobile's BLE (Bluetooth Low Energy) signal and Wi-Fi connection to AP (Access Point). If there is no user nearby, the TV intelligently turns off, to save energy (a manual-off timer is available too). To interact with both Android phone and iPhone, we design a general BLE advertisement for Android phone and an iBeacon format for iPhone. On the BLE proximity-based service, the Ambient Mode can be turned off/on based on the BLE signal presence. On the Wi-Fi connection based service for long-range detection, TV can detect mobile's AP connection using ARP (Address Resolution Protocol). The proposed service and intelligent controls have been implemented as real products and launched the service of Ambient Mode in the market.
{"title":"Ambient Mode: A Novel Service and Intelligent Control based on User Awareness using BLE and Wi-Fi","authors":"Suk-Un Yoon, Jinho Kim, Jongha Woo, Younghoon Moon, Cheul-hee Hahm","doi":"10.1109/icce-asia46551.2019.8942196","DOIUrl":"https://doi.org/10.1109/icce-asia46551.2019.8942196","url":null,"abstract":"In this paper, we introduce a novel service, Ambient Mode, which delivers a functional value to customers when TV is off. Instead of showing a meaningless black screen, the Ambient Mode provides meaningful and emotional background experiences. To mitigate users' concern about the cost of using the Ambient feature, the TV intelligently detects the presence of a person in the room from the registered mobile's BLE (Bluetooth Low Energy) signal and Wi-Fi connection to AP (Access Point). If there is no user nearby, the TV intelligently turns off, to save energy (a manual-off timer is available too). To interact with both Android phone and iPhone, we design a general BLE advertisement for Android phone and an iBeacon format for iPhone. On the BLE proximity-based service, the Ambient Mode can be turned off/on based on the BLE signal presence. On the Wi-Fi connection based service for long-range detection, TV can detect mobile's AP connection using ARP (Address Resolution Protocol). The proposed service and intelligent controls have been implemented as real products and launched the service of Ambient Mode in the market.","PeriodicalId":117814,"journal":{"name":"2019 IEEE International Conference on Consumer Electronics - Asia (ICCE-Asia)","volume":"70 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114831154","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 : 2019-06-01DOI: 10.1109/icce-asia46551.2019.8942221
Prasara Jakkaew, T. Onoye
Respiratory rate is the first observation to indicate a health problem. This study presents an approach to noncontact monitoring of respiratory rate and breathing pattern based on slow-motion images focus on sleeping positions. The movement while breathing is too tiny to be observed with the naked eyes. The body movement is captured by the slow-motion mode built in a smartphone camera. The primary benefit of this approach is the utilization of an accessibility device which everyone can use at home. The respiratory rate was obtained from the intensity value in the selected region of interest around the chest and abdomen area with used the Gaussian filter to reduce the noise. A motion tracking algorithm was implemented to track the region of interest movements. The obtained signal should be smoothed to reflect the breathing pattern then the Findpeaks function is applied in order to count the number of peaks for representing the number of the breaths. The result demonstrates that simple computer vision techniques can provide highly accurate breathing assessment. The accuracy depends on the location and size of region of interest, signal smoothing, and filter types. Besides, other variables affect accuracy, such as background views or patterns on clothing.
{"title":"An Approach to Non-contact Monitoring of Respiratory Rate and Breathing Pattern Based on Slow Motion Images","authors":"Prasara Jakkaew, T. Onoye","doi":"10.1109/icce-asia46551.2019.8942221","DOIUrl":"https://doi.org/10.1109/icce-asia46551.2019.8942221","url":null,"abstract":"Respiratory rate is the first observation to indicate a health problem. This study presents an approach to noncontact monitoring of respiratory rate and breathing pattern based on slow-motion images focus on sleeping positions. The movement while breathing is too tiny to be observed with the naked eyes. The body movement is captured by the slow-motion mode built in a smartphone camera. The primary benefit of this approach is the utilization of an accessibility device which everyone can use at home. The respiratory rate was obtained from the intensity value in the selected region of interest around the chest and abdomen area with used the Gaussian filter to reduce the noise. A motion tracking algorithm was implemented to track the region of interest movements. The obtained signal should be smoothed to reflect the breathing pattern then the Findpeaks function is applied in order to count the number of peaks for representing the number of the breaths. The result demonstrates that simple computer vision techniques can provide highly accurate breathing assessment. The accuracy depends on the location and size of region of interest, signal smoothing, and filter types. Besides, other variables affect accuracy, such as background views or patterns on clothing.","PeriodicalId":117814,"journal":{"name":"2019 IEEE International Conference on Consumer Electronics - Asia (ICCE-Asia)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129594234","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 : 2019-06-01DOI: 10.1109/icce-asia46551.2019.8942198
Rafee Al Ahsan, A. Baki
The world will see the standardization and deployment of 5G cellular technologies by the year 2020. Different modulation techniques are proposed in Internet of Things (IoT) based 5G, one of them is Universal Filtered Multi-Carrier (UFMC) system. UFMC uses Dolph-Chebyshev Filter to reduce the sub-band interferences. We have investigated a novel concept of Fractional Powered Binomial Filter (FPBF) for UFMC that can perform better than Dolph-Chebyshev Filter based UFMC. It was seen in our study that Dolph-Chebyshev Filter causes comparatively higher level of sub-band interference. This paper describes a better method of interference reduction among the sub-bands of UFMC-based 5G using novel FPBF. Bandwidth of each sub-band as well as adjacent-channel interference of UFMC can be easily controlled using a single parameter of FPBF.
{"title":"Implementation of Novel Fractional Powered Binomial Filter (FPBF) in 5G-UFMC","authors":"Rafee Al Ahsan, A. Baki","doi":"10.1109/icce-asia46551.2019.8942198","DOIUrl":"https://doi.org/10.1109/icce-asia46551.2019.8942198","url":null,"abstract":"The world will see the standardization and deployment of 5G cellular technologies by the year 2020. Different modulation techniques are proposed in Internet of Things (IoT) based 5G, one of them is Universal Filtered Multi-Carrier (UFMC) system. UFMC uses Dolph-Chebyshev Filter to reduce the sub-band interferences. We have investigated a novel concept of Fractional Powered Binomial Filter (FPBF) for UFMC that can perform better than Dolph-Chebyshev Filter based UFMC. It was seen in our study that Dolph-Chebyshev Filter causes comparatively higher level of sub-band interference. This paper describes a better method of interference reduction among the sub-bands of UFMC-based 5G using novel FPBF. Bandwidth of each sub-band as well as adjacent-channel interference of UFMC can be easily controlled using a single parameter of FPBF.","PeriodicalId":117814,"journal":{"name":"2019 IEEE International Conference on Consumer Electronics - Asia (ICCE-Asia)","volume":"24 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122249354","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 : 2019-06-01DOI: 10.1109/icce-asia46551.2019.8941599
M. V. Caya, Marvin U. Cosindad, Nicanor I. Marcelo, Jose Nicolas M. Santos, J. L. Torres
Most of the incidents involving intravenous infusion are attributed to the complexity of the administration process and insufficient medical service provider-to-patient ratio. The growing number incidents like these call for the development of an automated intravenous administration process. This paper describes the software aspect of an infusion control system for intravenous fluids including the development of a graphical user interface for infusion monitoring, creation of database for IV prescriptions and the automated flow control. This paper also compared the experimental drop rate as observed by the sensor with the manually obtained drop rate. In over 20 samples, the system produced a two-tailed p value of 0.4565 using a statistical hypothesis t-test.
{"title":"Design and Implementation of an Intravenous Infusion Control and Monitoring System","authors":"M. V. Caya, Marvin U. Cosindad, Nicanor I. Marcelo, Jose Nicolas M. Santos, J. L. Torres","doi":"10.1109/icce-asia46551.2019.8941599","DOIUrl":"https://doi.org/10.1109/icce-asia46551.2019.8941599","url":null,"abstract":"Most of the incidents involving intravenous infusion are attributed to the complexity of the administration process and insufficient medical service provider-to-patient ratio. The growing number incidents like these call for the development of an automated intravenous administration process. This paper describes the software aspect of an infusion control system for intravenous fluids including the development of a graphical user interface for infusion monitoring, creation of database for IV prescriptions and the automated flow control. This paper also compared the experimental drop rate as observed by the sensor with the manually obtained drop rate. In over 20 samples, the system produced a two-tailed p value of 0.4565 using a statistical hypothesis t-test.","PeriodicalId":117814,"journal":{"name":"2019 IEEE International Conference on Consumer Electronics - Asia (ICCE-Asia)","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117257695","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 : 2019-06-01DOI: 10.1109/icce-asia46551.2019.8942197
Anh H. Nguyen, Huyen T. T. Tran, Duc V. Nguyen, T. Thang
Current state-of-the-art deep learning-based models for human action recognition achieve impressive accuracy on benchmark datasets. However, the fact that those models are trained and tested on “clean” and high-quality input data raises a concern about their reliability under transmission artefacts and adversarial perturbations. In this work, we conduct for the first time an evaluation of the impacts of artefacts and adversarial attacks in deep learning-based human action recognition. Findings from this evaluation provide insights into the behaviors of action recognition under hostile conditions of best-effort networks.
{"title":"Impacts of Artefacts and Adversarial Attacks in Deep Learning based Action Recognition","authors":"Anh H. Nguyen, Huyen T. T. Tran, Duc V. Nguyen, T. Thang","doi":"10.1109/icce-asia46551.2019.8942197","DOIUrl":"https://doi.org/10.1109/icce-asia46551.2019.8942197","url":null,"abstract":"Current state-of-the-art deep learning-based models for human action recognition achieve impressive accuracy on benchmark datasets. However, the fact that those models are trained and tested on “clean” and high-quality input data raises a concern about their reliability under transmission artefacts and adversarial perturbations. In this work, we conduct for the first time an evaluation of the impacts of artefacts and adversarial attacks in deep learning-based human action recognition. Findings from this evaluation provide insights into the behaviors of action recognition under hostile conditions of best-effort networks.","PeriodicalId":117814,"journal":{"name":"2019 IEEE International Conference on Consumer Electronics - Asia (ICCE-Asia)","volume":"6 3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127692641","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}
Face occlusions with glasses or scarf are quite common in the real-world scenes, or more seriously, terrorists often cover their faces with sunglasses or a mask to hide themselves from the cameras. Occluded facial recognition is, therefore, an important problem in surveillance & defense department. A system that can recognize faces with occlusions may need to be trained by a huge set of facial databases. To reduce the complexity of an occluded facial recognition system, this paper investigates the effects of the two-dimensional principal component analysis (2DPCA) in the initialization phase on image classification by the convolutional neural network (CNN). Our experiments show that 2DPCA can reduce the image dimension for training while keeping the accuracy rate comparing to using the whole images. Our results, at 0.001 learning rate, showed 81.91% accuracy with 120 eigenvectors for the AR database, and 99.95 % accuracy rate with 190 eigenvectors for the GTAV database.
{"title":"Occluded Facial Recognition with 2DPCA based Convolutional Neural Network","authors":"Sittiphan Sarapakdi, Phaderm Nangsue, Charnchai Pluempitiwirivawej","doi":"10.1109/icce-asia46551.2019.8942204","DOIUrl":"https://doi.org/10.1109/icce-asia46551.2019.8942204","url":null,"abstract":"Face occlusions with glasses or scarf are quite common in the real-world scenes, or more seriously, terrorists often cover their faces with sunglasses or a mask to hide themselves from the cameras. Occluded facial recognition is, therefore, an important problem in surveillance & defense department. A system that can recognize faces with occlusions may need to be trained by a huge set of facial databases. To reduce the complexity of an occluded facial recognition system, this paper investigates the effects of the two-dimensional principal component analysis (2DPCA) in the initialization phase on image classification by the convolutional neural network (CNN). Our experiments show that 2DPCA can reduce the image dimension for training while keeping the accuracy rate comparing to using the whole images. Our results, at 0.001 learning rate, showed 81.91% accuracy with 120 eigenvectors for the AR database, and 99.95 % accuracy rate with 190 eigenvectors for the GTAV database.","PeriodicalId":117814,"journal":{"name":"2019 IEEE International Conference on Consumer Electronics - Asia (ICCE-Asia)","volume":"438 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115936714","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：