Zhe Huang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Zhe Huang has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
- Chengan Guo
Dalian University of Technology
2 shared publications - Benjamin S. Wessler
Tufts Clinical and Translational Science Institute, Tufts Medical Center, Tufts University , Boston, MA
1 shared publication - Changqing Zou
Zhejiang Lab, Hangzhou, China and Zhejiang University, Hangzhou, China
1 shared publication - Hongbo Fu
Hong Kong University of Science and Technology, Hong Kong, Hong Kong
1 shared publication - Jianzhuang Liu
Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China
1 shared publication - Likun Tang
Chengdu University of Technology,School of Electrical and Mechanical Engineering,Chengdu,Sichuan,China
1 shared publication
Latest Publications
- 1 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-12
Toward Cross-Dataset Finger Vein Recognition With Single-Source Data
Finger vein recognition (FVR) has recently gained significant attention due to its high security, contactless acquisition, and reliability. However, the principle of vein imaging and touchless mode makes vein images susceptible to variations caused by different acquisition devices and environmental conditions, resulting in gaps among data from different sources. As recognition systems need to be deployed on various devices under uncontrolled environments, FVR faces the challenge of cross-dataset recognition. Moreover, sharing data from multiple datasets is often infeasible due to privacy and security concerns, leading to that training can only be performed using a single-source dataset. To address cross-dataset FVR with single-source data, this article proposes a comprehensive approach. First, an instance-based adaptive transformation component is introduced to reduce dataset gaps at the instance level by removing uncontrolled bias fields that cause dataset discrepancies, adjusting the latent distribution to be more consistent and enhancing vein patterns. Second, a feature alignment and clustering learning algorithm is employed at the feature level to learn invariant features against intraclass variations and perturbations. Third, to fast deploy the deep model to different target datasets with anticipated recognition performance, a modified extreme learning machine (ELM)-based classifier is adopted as a task-specific classifier to quicken the learning speed and improve the generalization capability of the model. Extensive experiments are conducted on four public finger vein datasets, demonstrating the superiority of the proposed method in addressing cross-dataset FVR challenges.
- 1 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-12
Toward Cross-Dataset Finger Vein Recognition With Single-Source Data
Finger vein recognition (FVR) has recently gained significant attention due to its high security, contactless acquisition, and reliability. However, the principle of vein imaging and touchless mode makes vein images susceptible to variations caused by different acquisition devices and environmental conditions, resulting in gaps among data from different sources. As recognition systems need to be deployed on various devices under uncontrolled environments, FVR faces the challenge of cross-dataset recognition. Moreover, sharing data from multiple datasets is often infeasible due to privacy and security concerns, leading to that training can only be performed using a single-source dataset. To address cross-dataset FVR with single-source data, this article proposes a comprehensive approach. First, an instance-based adaptive transformation component is introduced to reduce dataset gaps at the instance level by removing uncontrolled bias fields that cause dataset discrepancies, adjusting the latent distribution to be more consistent and enhancing vein patterns. Second, a feature alignment and clustering learning algorithm is employed at the feature level to learn invariant features against intraclass variations and perturbations. Third, to fast deploy the deep model to different target datasets with anticipated recognition performance, a modified extreme learning machine (ELM)-based classifier is adopted as a task-specific classifier to quicken the learning speed and improve the generalization capability of the model. Extensive experiments are conducted on four public finger vein datasets, demonstrating the superiority of the proposed method in addressing cross-dataset FVR challenges.
- 8 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 17, 409-413
Domain Adaptive Recognition of Defects for Ground Penetrating Radar Images
GPR technology is of significant importance for road maintenance authorities to detect road defect under the road surface, for it helps to monitor the construction process and detect defects early, thus reducing maintenance costs. The high cost of data collection and core verification for underground defects makes the dataset very valuable. In practical applications, different radar types and imaging methods often lead to heterogeneous generated radar images. In addition, there are significant differences between training data and real-world test data, which may lead to ambiguous predictions. This is because as the discrepance in different data domains increase, the transferability of the features extracted by neural networks significantly decreases at higher levels. Therefore, it is important to reduce dataset bias and enhance transferability in specific task layers. This article specifically designs a feature transfer enhancement algorithm for heterogeneous ground penetrating radar data based on deep domain adaptation. Compared with other deep domain adaptation algorithms such as DAN, DANN, and DeepCoral, the proposed method has the best recognition performance on underground defect data of ground penetrating radar.
- 8 July 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 17, 409-413
Domain Adaptive Recognition of Defects for Ground Penetrating Radar Images
GPR technology is of significant importance for road maintenance authorities to detect road defect under the road surface, for it helps to monitor the construction process and detect defects early, thus reducing maintenance costs. The high cost of data collection and core verification for underground defects makes the dataset very valuable. In practical applications, different radar types and imaging methods often lead to heterogeneous generated radar images. In addition, there are significant differences between training data and real-world test data, which may lead to ambiguous predictions. This is because as the discrepance in different data domains increase, the transferability of the features extracted by neural networks significantly decreases at higher levels. Therefore, it is important to reduce dataset bias and enhance transferability in specific task layers. This article specifically designs a feature transfer enhancement algorithm for heterogeneous ground penetrating radar data based on deep domain adaptation. Compared with other deep domain adaptation algorithms such as DAN, DANN, and DeepCoral, the proposed method has the best recognition performance on underground defect data of ground penetrating radar.
- Open Access
Fix-A-Step: Semi-supervised Learning from Uncurated Unlabeled Data
Semi-supervised learning (SSL) promises improved accuracy compared to training classifiers on small labeled datasets by also training on many unlabeled images. In real applications like medical imaging, unlabeled data will be collected for expediency and thus uncurated: possibly different from the labeled set in classes or features. Unfortunately, modern deep SSL often makes accuracy worse when given uncurated unlabeled data. Recent complex remedies try to detect out-of-distribution unlabeled images and then discard or downweight them. Instead, we introduce Fix-A-Step, a simpler procedure that views all uncurated unlabeled images as potentially helpful. Our first insight is that even uncurated images can yield useful augmentations of labeled data. Second, we modify gradient descent updates to prevent optimizing a multi-task SSL loss from hurting labeled-set accuracy. Fix-A-Step can repair many common deep SSL methods, improving accuracy on CIFAR benchmarks across all tested methods and levels of artificial class mismatch. On a new medical SSL benchmark called Heart2Heart, Fix-A-Step can learn from 353,500 truly uncurated ultrasound images to deliver gains that generalize across hospitals.
Open AccessFix-A-Step: Semi-supervised Learning from Uncurated Unlabeled Data
Semi-supervised learning (SSL) promises improved accuracy compared to training classifiers on small labeled datasets by also training on many unlabeled images. In real applications like medical imaging, unlabeled data will be collected for expediency and thus uncurated: possibly different from the labeled set in classes or features. Unfortunately, modern deep SSL often makes accuracy worse when given uncurated unlabeled data. Recent complex remedies try to detect out-of-distribution unlabeled images and then discard or downweight them. Instead, we introduce Fix-A-Step, a simpler procedure that views all uncurated unlabeled images as potentially helpful. Our first insight is that even uncurated images can yield useful augmentations of labeled data. Second, we modify gradient descent updates to prevent optimizing a multi-task SSL loss from hurting labeled-set accuracy. Fix-A-Step can repair many common deep SSL methods, improving accuracy on CIFAR benchmarks across all tested methods and levels of artificial class mismatch. On a new medical SSL benchmark called Heart2Heart, Fix-A-Step can learn from 353,500 truly uncurated ultrasound images to deliver gains that generalize across hospitals.
- Open Access
- 1 January 2021
- Published by Wiley in Computational Intelligence and Neuroscience
- Vol. 2021 (1)
Automatic Diagnosis of Alzheimer’s Disease and Mild Cognitive Impairment Based on CNN + SVM Networks with End‐to‐End Training
Alzheimer’s disease (AD) is an irreversible neurodegenerative disease, and, at present, once it has been diagnosed, there is no effective curative treatment. Accurate and early diagnosis of Alzheimer’s disease is crucial for improving the condition of patients since effective preventive measures can be taken in advance to delay the onset time of the disease. 18F‐Fluorodeoxyglucose positron emission tomography (18F‐FDG PET : PET) is an effective biomarker of the symptom of AD and has been used as medical imaging data for diagnosing AD. Mild cognitive impairment (MCI) is regarded as an early symptom of AD, and it has been shown that MCI also has a certain biomedical correlation with PET. In this paper, we explore how to use 3D PET images to realize the effective recognition of MCI and thus achieve the early prediction of AD. This problem is then taken as the classification of three categories of PET images, including MCI, AD, and NC (normal controls). In order to get better classification performance, a novel network model is proposed in the paper based on 3D convolution neural networks (CNN) and support vector machines (SVM) by utilizing both the excellent abilities of CNN in feature extraction and SVM in classification. In order to make full use of the optimal property of SVM in solving binary classification problems, the three‐category classification problem is divided into three binary classifications, and each binary classification is being realized with a CNN + SVM network. Then, the outputs of the three CNN + SVM networks are fused into a final three‐category classification result. An end‐to‐end learning algorithm is developed to train the CNN + SVM networks, and a decision fusion algorithm is exploited to realize the fusion of the outputs of three CNN + SVM networks. Experimental results obtained in the work with comparative analyses confirm the effectiveness of the proposed method.
- 1 January 2021
- Published by Wiley in Computational Intelligence and Neuroscience
- Vol. 2021 (1)
Open AccessAutomatic Diagnosis of Alzheimer’s Disease and Mild Cognitive Impairment Based on CNN + SVM Networks with End‐to‐End Training
Alzheimer’s disease (AD) is an irreversible neurodegenerative disease, and, at present, once it has been diagnosed, there is no effective curative treatment. Accurate and early diagnosis of Alzheimer’s disease is crucial for improving the condition of patients since effective preventive measures can be taken in advance to delay the onset time of the disease. 18F‐Fluorodeoxyglucose positron emission tomography (18F‐FDG PET : PET) is an effective biomarker of the symptom of AD and has been used as medical imaging data for diagnosing AD. Mild cognitive impairment (MCI) is regarded as an early symptom of AD, and it has been shown that MCI also has a certain biomedical correlation with PET. In this paper, we explore how to use 3D PET images to realize the effective recognition of MCI and thus achieve the early prediction of AD. This problem is then taken as the classification of three categories of PET images, including MCI, AD, and NC (normal controls). In order to get better classification performance, a novel network model is proposed in the paper based on 3D convolution neural networks (CNN) and support vector machines (SVM) by utilizing both the excellent abilities of CNN in feature extraction and SVM in classification. In order to make full use of the optimal property of SVM in solving binary classification problems, the three‐category classification problem is divided into three binary classifications, and each binary classification is being realized with a CNN + SVM network. Then, the outputs of the three CNN + SVM networks are fused into a final three‐category classification result. An end‐to‐end learning algorithm is developed to train the CNN + SVM networks, and a decision fusion algorithm is exploited to realize the fusion of the outputs of three CNN + SVM networks. Experimental results obtained in the work with comparative analyses confirm the effectiveness of the proposed method.
- Open Access
- 21 April 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 84230-84238
Secure Captchas via Object Segment Collages
CAPTCHA, a short form of “Completely Automated Public Turing test to Tell Computers and Humans Apart”, is a computer program that can generate and grade tests that most humans can pass, but current computer programs cannot. This paper advances research on image-based CAPTCHA by incorporating the object segment collages. We first collect object segments from thousands of images by using our self-developed collection tool. Then, we create an image collage that compiles randomly selected object segments. To increase the difficulty for computer detection algorithms, we embed some features such as noisy background, object segment occlusion and object segment orientation update. We also develop a practical application using the newly proposed Object Segment Collage CAPTCHA. For the evaluation, we measure the accuracy performance of computer vs. human in solving CAPTCHA questions. In addition, we also conduct a user study over different CAPTCHA methods.
- 21 April 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 84230-84238
Open AccessSecure Captchas via Object Segment Collages
CAPTCHA, a short form of “Completely Automated Public Turing test to Tell Computers and Humans Apart”, is a computer program that can generate and grade tests that most humans can pass, but current computer programs cannot. This paper advances research on image-based CAPTCHA by incorporating the object segment collages. We first collect object segments from thousands of images by using our self-developed collection tool. Then, we create an image collage that compiles randomly selected object segments. To increase the difficulty for computer detection algorithms, we embed some features such as noisy background, object segment occlusion and object segment orientation update. We also develop a practical application using the newly proposed Object Segment Collage CAPTCHA. For the evaluation, we measure the accuracy performance of computer vs. human in solving CAPTCHA questions. In addition, we also conduct a user study over different CAPTCHA methods.
- Open Access
Sketch-based Shape Retrieval using Pyramid-of-Parts
We present a multi-scale approach to sketch-based shape retrieval. It is based on a novel multi-scale shape descriptor called Pyramidof- Parts, which encodes the features and spatial relationship of the semantic parts of query sketches. The same descriptor can also be used to represent 2D projected views of 3D shapes, allowing effective matching of query sketches with 3D shapes across multiple scales. Experimental results show that the proposed method outperforms the state-of-the-art method, whether the sketch segmentation information is obtained manually or automatically by considering each stroke as a semantic part.
Open AccessSketch-based Shape Retrieval using Pyramid-of-Parts
We present a multi-scale approach to sketch-based shape retrieval. It is based on a novel multi-scale shape descriptor called Pyramidof- Parts, which encodes the features and spatial relationship of the semantic parts of query sketches. The same descriptor can also be used to represent 2D projected views of 3D shapes, allowing effective matching of query sketches with 3D shapes across multiple scales. Experimental results show that the proposed method outperforms the state-of-the-art method, whether the sketch segmentation information is obtained manually or automatically by considering each stroke as a semantic part.