Pub Date : 2018-11-01DOI: 10.1109/ICDSP.2018.8631801
Yuechi Jiang, F. H. F. Leung
Fisher Discriminant Analysis (FDA) is a widely used technique for signal classification. Its application varies from face recognition to speaker recognition. FDA aims to project a given feature onto a projected space, where the features coming from the same class are moved closer, while those coming from different classes are moved farther. However, in the original formulation of FDA, the number of orthogonal projection directions is limited by the number of classes, which may hinder the effectiveness of FDA as a projection technique. In this paper, we propose to use new between-class scatter matrices to replace the original between-class scatter matrix, in order to increase the number of orthogonal projection directions. We call FDA with these new between-class scatter matrices the Modified FDA (MFDA). The effectiveness of MFDA and FDA as a projection technique is compared through doing two audio signal classification tasks. Both linear version and kernel version of MFDA and FDA are evaluated, and experimental results show that MFDA can outperform FDA in both classification tasks.
{"title":"Fisher Discriminant Analysis with New Between-class Scatter Matrix for Audio Signal Classification","authors":"Yuechi Jiang, F. H. F. Leung","doi":"10.1109/ICDSP.2018.8631801","DOIUrl":"https://doi.org/10.1109/ICDSP.2018.8631801","url":null,"abstract":"Fisher Discriminant Analysis (FDA) is a widely used technique for signal classification. Its application varies from face recognition to speaker recognition. FDA aims to project a given feature onto a projected space, where the features coming from the same class are moved closer, while those coming from different classes are moved farther. However, in the original formulation of FDA, the number of orthogonal projection directions is limited by the number of classes, which may hinder the effectiveness of FDA as a projection technique. In this paper, we propose to use new between-class scatter matrices to replace the original between-class scatter matrix, in order to increase the number of orthogonal projection directions. We call FDA with these new between-class scatter matrices the Modified FDA (MFDA). The effectiveness of MFDA and FDA as a projection technique is compared through doing two audio signal classification tasks. Both linear version and kernel version of MFDA and FDA are evaluated, and experimental results show that MFDA can outperform FDA in both classification tasks.","PeriodicalId":218806,"journal":{"name":"2018 IEEE 23rd International Conference on Digital Signal Processing (DSP)","volume":"40 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123954818","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 : 2018-11-01DOI: 10.1109/ICDSP.2018.8631677
Ran Zhu, Zhuoling Xiao, Mo Cheng, Liang Zhou, Bo Yan, Shuisheng Lin, Hongkai Wen
The ubiquity of smartphones and their rich set of onboard sensors have created many exciting new opportunities. One important application is activity recognition based on smartphone inertial sensors, which is a fundamental building block for a variety of scenarios, such as indoor pedestrian tracking, mobile health care and smart cities. Though many approaches have been proposed to address the human activity recognition problem, a number of challenges still present: (i) people’s motion modes are very different; (ii) there is very limited amount of training data; (iii) human activities can be arbitrary and complex, and thus handcrafted feature engineering often fail to work; and finally (iv) the recognition accuracy tends to be limited due to confusing activities. To tackle those challenges, in this paper we propose a human activity recognition framework based on Convolutional Neural Network (CNN) using smartphone-based accelerometer, gyroscope, and magnetometer, which achieves 95.62% accuracy, and also presents a novel ensembles of CNN solving the confusion between certain activities like going upstairs and walking. Extensive experiments have been conducted using 153088 sensory samples from 100 subjects. The results show that the classification accuracy of the generalized model can reach 96.29%.
{"title":"Deep Ensemble Learning for Human Activity Recognition Using Smartphone","authors":"Ran Zhu, Zhuoling Xiao, Mo Cheng, Liang Zhou, Bo Yan, Shuisheng Lin, Hongkai Wen","doi":"10.1109/ICDSP.2018.8631677","DOIUrl":"https://doi.org/10.1109/ICDSP.2018.8631677","url":null,"abstract":"The ubiquity of smartphones and their rich set of onboard sensors have created many exciting new opportunities. One important application is activity recognition based on smartphone inertial sensors, which is a fundamental building block for a variety of scenarios, such as indoor pedestrian tracking, mobile health care and smart cities. Though many approaches have been proposed to address the human activity recognition problem, a number of challenges still present: (i) people’s motion modes are very different; (ii) there is very limited amount of training data; (iii) human activities can be arbitrary and complex, and thus handcrafted feature engineering often fail to work; and finally (iv) the recognition accuracy tends to be limited due to confusing activities. To tackle those challenges, in this paper we propose a human activity recognition framework based on Convolutional Neural Network (CNN) using smartphone-based accelerometer, gyroscope, and magnetometer, which achieves 95.62% accuracy, and also presents a novel ensembles of CNN solving the confusion between certain activities like going upstairs and walking. Extensive experiments have been conducted using 153088 sensory samples from 100 subjects. The results show that the classification accuracy of the generalized model can reach 96.29%.","PeriodicalId":218806,"journal":{"name":"2018 IEEE 23rd International Conference on Digital Signal Processing (DSP)","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124108545","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 : 2018-11-01DOI: 10.1109/ICDSP.2018.8631561
Liping Chang, Jianjun Yang, Sheng Li, Hong Xu, Kai Liu, Chaogeng Huang
Face recognition is one of the most challenging topics in the field of machine vision and pattern recognition, and has a wide range of applications. The face features play an important role in the classification, while the features extracted by traditional methods are simple and elementary. To solve this problem, a stacked convolutional autoencoder (SCAE) based on deep learning theory is used to extract deeper features. The output of the encoder can be taken to design a feature dictionary. Meanwhile sparse representation is a general classification algorithm which has shown the good performance in the field of object recognition. In this paper a framework based on stacked convolutional autoencoder and sparse representation is proposed. Experiments, carried out with the LFW face database, have shown that the proposed framework can extract more deep and abstract features by multi-level cascade, and has high recognition speed and high accuracy.
{"title":"Face Recognition Based on Stacked Convolutional Autoencoder and Sparse Representation","authors":"Liping Chang, Jianjun Yang, Sheng Li, Hong Xu, Kai Liu, Chaogeng Huang","doi":"10.1109/ICDSP.2018.8631561","DOIUrl":"https://doi.org/10.1109/ICDSP.2018.8631561","url":null,"abstract":"Face recognition is one of the most challenging topics in the field of machine vision and pattern recognition, and has a wide range of applications. The face features play an important role in the classification, while the features extracted by traditional methods are simple and elementary. To solve this problem, a stacked convolutional autoencoder (SCAE) based on deep learning theory is used to extract deeper features. The output of the encoder can be taken to design a feature dictionary. Meanwhile sparse representation is a general classification algorithm which has shown the good performance in the field of object recognition. In this paper a framework based on stacked convolutional autoencoder and sparse representation is proposed. Experiments, carried out with the LFW face database, have shown that the proposed framework can extract more deep and abstract features by multi-level cascade, and has high recognition speed and high accuracy.","PeriodicalId":218806,"journal":{"name":"2018 IEEE 23rd International Conference on Digital Signal Processing (DSP)","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114019545","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}
Dynamic magnetic resonance imaging (DMRI) sequence can be represented as the sum of a low-rank component and a sparse tensor component. To exploit the low rank structure in multi-way data, the current works use either the Tucker rank or the CANDECOMP/PARAFAC (CP) rank for the low rank tensor component. In fact, these two kinds of tensor ranks represent different structures in high-dimensional data. In this paper, We propose a multiple low ranks plus sparsity based tensor reconstruction method for DMRI. The simultaneous minimization of both CP and Tucker ranks can better exploit multi-dimensional coherence in the low rank component of DMRI data, and the sparse component is regularized by the tensor total variation minimization. The reconstruction optimization model can be divided into two sub-problems to iteratively calculate the low rank and sparse components. For the sub-problem about low rank tensor component, the rank-one tensor updating and sum of nuclear norm minimization methods are used to solve it. To obtain the sparse tensor component, the primal dual method is used. We compare the proposed method with four state-of-the-art ones, and experimental results show that the proposed method can achieve better reconstruction quality than state-of-the-art ones.
{"title":"Multiple Low-Ranks plus Sparsity based Tensor Reconstruction for Dynamic MRI","authors":"Shan Wu, Yipeng Liu, Tengteng Liu, Fei Wen, Sayuan Liang, Xiang Zhang, Shuai Wang, Ce Zhu","doi":"10.1109/ICDSP.2018.8631646","DOIUrl":"https://doi.org/10.1109/ICDSP.2018.8631646","url":null,"abstract":"Dynamic magnetic resonance imaging (DMRI) sequence can be represented as the sum of a low-rank component and a sparse tensor component. To exploit the low rank structure in multi-way data, the current works use either the Tucker rank or the CANDECOMP/PARAFAC (CP) rank for the low rank tensor component. In fact, these two kinds of tensor ranks represent different structures in high-dimensional data. In this paper, We propose a multiple low ranks plus sparsity based tensor reconstruction method for DMRI. The simultaneous minimization of both CP and Tucker ranks can better exploit multi-dimensional coherence in the low rank component of DMRI data, and the sparse component is regularized by the tensor total variation minimization. The reconstruction optimization model can be divided into two sub-problems to iteratively calculate the low rank and sparse components. For the sub-problem about low rank tensor component, the rank-one tensor updating and sum of nuclear norm minimization methods are used to solve it. To obtain the sparse tensor component, the primal dual method is used. We compare the proposed method with four state-of-the-art ones, and experimental results show that the proposed method can achieve better reconstruction quality than state-of-the-art ones.","PeriodicalId":218806,"journal":{"name":"2018 IEEE 23rd International Conference on Digital Signal Processing (DSP)","volume":"101 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124057664","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 : 2018-11-01DOI: 10.1109/ICDSP.2018.8631602
M. A. Afzal, Di He, Ziyu Zhu, Yueming Yang
In the period of wireless communication, Indoor positioning systems (IPSs) are getting enormous attention. These systems are construct to attain location information of individuals and objects inside a building. Now a day all the applicable wireless technologies used in this context are Wi-Fi and Bluetooth Low Energy (BLE) based. These advancements are additionally decided for their ease of use, low cost and integration into wireless devices. However, these techniques are having some positioning errors along with specified region. Here, we introduce another wireless technology named Light Fidelity (Li-Fi), the basic convention of this technology is the transfer of information using light illumination by light emitting diodes. This article primarily established a set of evaluation indexes for the performance of these three Wi-Fi, BLE and Li-Fi technologies in indoor positioning scenarios. We compare the predefined IPSs in term of performance and limitations. After then outline the tradeoffs among these systems from the perspective of all evaluation entities. We show experimentally that Li-Fi technology achieves a high efficiency but for accuracy, Wi-Fi is still better than Li-Fi and BLE technology.
{"title":"Performance Evaluation of Wi-Fi Bluetooth Low Energy & Li-Fi Technology in Indoor Positioning","authors":"M. A. Afzal, Di He, Ziyu Zhu, Yueming Yang","doi":"10.1109/ICDSP.2018.8631602","DOIUrl":"https://doi.org/10.1109/ICDSP.2018.8631602","url":null,"abstract":"In the period of wireless communication, Indoor positioning systems (IPSs) are getting enormous attention. These systems are construct to attain location information of individuals and objects inside a building. Now a day all the applicable wireless technologies used in this context are Wi-Fi and Bluetooth Low Energy (BLE) based. These advancements are additionally decided for their ease of use, low cost and integration into wireless devices. However, these techniques are having some positioning errors along with specified region. Here, we introduce another wireless technology named Light Fidelity (Li-Fi), the basic convention of this technology is the transfer of information using light illumination by light emitting diodes. This article primarily established a set of evaluation indexes for the performance of these three Wi-Fi, BLE and Li-Fi technologies in indoor positioning scenarios. We compare the predefined IPSs in term of performance and limitations. After then outline the tradeoffs among these systems from the perspective of all evaluation entities. We show experimentally that Li-Fi technology achieves a high efficiency but for accuracy, Wi-Fi is still better than Li-Fi and BLE technology.","PeriodicalId":218806,"journal":{"name":"2018 IEEE 23rd International Conference on Digital Signal Processing (DSP)","volume":"42 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127668195","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 : 2018-11-01DOI: 10.1109/ICDSP.2018.8631659
H. Kwan, Jiajun Liang, A. Jiang
In this paper, sparse linear phase FIR Iowpass digital filter design using iterative cuckoo search algorithm with step-descendant coefficient thresholding is presented. During each iteration, the least-squares frequency response error is minimized using cuckoo search algorithm. A step-descendant coefficient threshold is used to iteratively update the zero-valued filter coefficients. With the same set of sparsity levels, the obtained design results indicate that smaller weighted least-squares errors and slightly smaller peak magnitude errors can be obtained when compared to those of a recent design method.
{"title":"Sparse Linear Phase FIR Filter Design using Iterative CSA","authors":"H. Kwan, Jiajun Liang, A. Jiang","doi":"10.1109/ICDSP.2018.8631659","DOIUrl":"https://doi.org/10.1109/ICDSP.2018.8631659","url":null,"abstract":"In this paper, sparse linear phase FIR Iowpass digital filter design using iterative cuckoo search algorithm with step-descendant coefficient thresholding is presented. During each iteration, the least-squares frequency response error is minimized using cuckoo search algorithm. A step-descendant coefficient threshold is used to iteratively update the zero-valued filter coefficients. With the same set of sparsity levels, the obtained design results indicate that smaller weighted least-squares errors and slightly smaller peak magnitude errors can be obtained when compared to those of a recent design method.","PeriodicalId":218806,"journal":{"name":"2018 IEEE 23rd International Conference on Digital Signal Processing (DSP)","volume":"3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121789760","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 : 2018-11-01DOI: 10.1109/ICDSP.2018.8631704
Amby Mao
Cancer becomes No.1 killing disease in China now. The paper describes how to generate virtual tumor signals by mathematical modeling, how to deeply process the cancer signals in chemotherapy, radiotherapy, target therapy and bioimmunotherapy by AI algorithms and how to design an AI chip in nano-drug delivery system for lung cancer. The purpose for this paper is to change the straightforward rules-based treatment guidelines about one drug fitting all and one dose fitting all in traditional cancer treatments into precision and personalized cancer treatments with advanced artificial intelligent technology. We hope the state of art in this technology could prolong and improve the cancer patient’s life and quality, let cancer become chronic disease in near future.
{"title":"Cancer Signals in Deep Processing","authors":"Amby Mao","doi":"10.1109/ICDSP.2018.8631704","DOIUrl":"https://doi.org/10.1109/ICDSP.2018.8631704","url":null,"abstract":"Cancer becomes No.1 killing disease in China now. The paper describes how to generate virtual tumor signals by mathematical modeling, how to deeply process the cancer signals in chemotherapy, radiotherapy, target therapy and bioimmunotherapy by AI algorithms and how to design an AI chip in nano-drug delivery system for lung cancer. The purpose for this paper is to change the straightforward rules-based treatment guidelines about one drug fitting all and one dose fitting all in traditional cancer treatments into precision and personalized cancer treatments with advanced artificial intelligent technology. We hope the state of art in this technology could prolong and improve the cancer patient’s life and quality, let cancer become chronic disease in near future.","PeriodicalId":218806,"journal":{"name":"2018 IEEE 23rd International Conference on Digital Signal Processing (DSP)","volume":"34 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131994840","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 : 2018-11-01DOI: 10.1109/ICDSP.2018.8631789
Lang Zou, Xiaofeng Liu, A. Jiang, Xu Zhou
In this paper, a patient specific seizure detection system using channel-restricted convolutional neural network(CR-CNN) with deep structure is represented. The binary patterns of brainwave activity reflected on ictal and interictal EEG are auto-memorized based on back-propagation mechanism. It is well trained using massive historical scalp EEG data of 23 pediatric patients with epilepsy from CHB-MIT database. Experimental results demonstrate that the proposed detector achieves the state of the art performance. The average false alarms rate reaches 0.12 per hour and only one out of the 167 seizures is missed.
{"title":"Epileptic Seizure Detection Using Deep Convolutional Network","authors":"Lang Zou, Xiaofeng Liu, A. Jiang, Xu Zhou","doi":"10.1109/ICDSP.2018.8631789","DOIUrl":"https://doi.org/10.1109/ICDSP.2018.8631789","url":null,"abstract":"In this paper, a patient specific seizure detection system using channel-restricted convolutional neural network(CR-CNN) with deep structure is represented. The binary patterns of brainwave activity reflected on ictal and interictal EEG are auto-memorized based on back-propagation mechanism. It is well trained using massive historical scalp EEG data of 23 pediatric patients with epilepsy from CHB-MIT database. Experimental results demonstrate that the proposed detector achieves the state of the art performance. The average false alarms rate reaches 0.12 per hour and only one out of the 167 seizures is missed.","PeriodicalId":218806,"journal":{"name":"2018 IEEE 23rd International Conference on Digital Signal Processing (DSP)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130217166","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 : 2018-11-01DOI: 10.1109/ICDSP.2018.8631819
Songze Zhang, Junjie Xie, Hongjian Shi
Nowadays, more people pay attention to the dental health including oral cavities, bone tumors or cancers, so the dental CBCT images becomes popular and are widely used in dental diagnosis. Dental implants, orthodontic orthodontics and other surgical procedures are employed in daily life. Accurate jaw separation from neighboring tissues can greatly improve diagnosis results, space measurements and success rates of surgical operations. This paper proposes an automatic segmentation algorithm to separate jaw bone from CBCT images. This algorithm uses the idea of three-dimensional region growing to perform segmentation, then optimizes the segmentation results with active contours. This algorithm yields more accurate segmentation of the jaw bone. Experiments are performed to both manually and automatically segment 10 groups of CBCT datasets. With manual segmentation references, our algorithm demonstrated our automatic segmentation algorithm work well, and further confirmed by evaluation of four quantitative metrics PSNR, SSIM, Precision and Recall. It can potentially assist doctors in diagnosis and surgical planning.
{"title":"Jaw Segmentation from CBCT Images","authors":"Songze Zhang, Junjie Xie, Hongjian Shi","doi":"10.1109/ICDSP.2018.8631819","DOIUrl":"https://doi.org/10.1109/ICDSP.2018.8631819","url":null,"abstract":"Nowadays, more people pay attention to the dental health including oral cavities, bone tumors or cancers, so the dental CBCT images becomes popular and are widely used in dental diagnosis. Dental implants, orthodontic orthodontics and other surgical procedures are employed in daily life. Accurate jaw separation from neighboring tissues can greatly improve diagnosis results, space measurements and success rates of surgical operations. This paper proposes an automatic segmentation algorithm to separate jaw bone from CBCT images. This algorithm uses the idea of three-dimensional region growing to perform segmentation, then optimizes the segmentation results with active contours. This algorithm yields more accurate segmentation of the jaw bone. Experiments are performed to both manually and automatically segment 10 groups of CBCT datasets. With manual segmentation references, our algorithm demonstrated our automatic segmentation algorithm work well, and further confirmed by evaluation of four quantitative metrics PSNR, SSIM, Precision and Recall. It can potentially assist doctors in diagnosis and surgical planning.","PeriodicalId":218806,"journal":{"name":"2018 IEEE 23rd International Conference on Digital Signal Processing (DSP)","volume":"55 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133915120","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 : 2018-11-01DOI: 10.1109/ICDSP.2018.8631568
Chuan-Yu Chang, Man-Ju Cheng, M. Ma
According to clinical reports, people with ages between 60 to 79 years have a high risk of stroke. The most obvious facial features of stroke are expressional asymmetry and mouth askew. In this study, we proposed a facial stroke recognition system that assists patients in self-judgment. Facial landmarks were tracked by an ensemble of regression trees (ERT) method. Two symmetry indexes area ratio and distance ratio between the left and right side of the eye and mouth were calculated. Local Ternary Pattern (LTP) and Gabor filter were used to enhance and to extract the texture features of the region of interest (ROI), respectively. The structural similarity of ROI between the left and right face was calculated. After that, we modified the original feature selection algorithm to select the best feature set. To classify facial stroke, the Support Vector Machine (SVM), Random Forest (RF), and Bayesian Classifier were adopted as classifier. The experimental results show that the proposed system can accurately and effectively distinguish stroke from facial images. The recognition accuracy of SVM, Random Forest, and Bayes are 100%, 95.45%, and 100%, respectively.
{"title":"Application of Machine Learning for Facial Stroke Detection","authors":"Chuan-Yu Chang, Man-Ju Cheng, M. Ma","doi":"10.1109/ICDSP.2018.8631568","DOIUrl":"https://doi.org/10.1109/ICDSP.2018.8631568","url":null,"abstract":"According to clinical reports, people with ages between 60 to 79 years have a high risk of stroke. The most obvious facial features of stroke are expressional asymmetry and mouth askew. In this study, we proposed a facial stroke recognition system that assists patients in self-judgment. Facial landmarks were tracked by an ensemble of regression trees (ERT) method. Two symmetry indexes area ratio and distance ratio between the left and right side of the eye and mouth were calculated. Local Ternary Pattern (LTP) and Gabor filter were used to enhance and to extract the texture features of the region of interest (ROI), respectively. The structural similarity of ROI between the left and right face was calculated. After that, we modified the original feature selection algorithm to select the best feature set. To classify facial stroke, the Support Vector Machine (SVM), Random Forest (RF), and Bayesian Classifier were adopted as classifier. The experimental results show that the proposed system can accurately and effectively distinguish stroke from facial images. The recognition accuracy of SVM, Random Forest, and Bayes are 100%, 95.45%, and 100%, respectively.","PeriodicalId":218806,"journal":{"name":"2018 IEEE 23rd International Conference on Digital Signal Processing (DSP)","volume":"53 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128945895","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：