Pub Date : 2018-12-01DOI: 10.1109/DICTA.2018.8615765
Heng Wang, Donghao Zhang, Yang Song, Siqi Liu, Rong Gao, Hanchuan Peng, Weidong (Tom) Cai
3D reconstruction of neuronal morphology is crucial to solving neuron-related problems in neuroscience as it is a key technique for investigating the connectivity and functionality of the neuron system. Many methods have been proposed to improve the accuracy of digital neuron reconstruction. However, the large amount of computer memory and computation time they require to process the large-scale images have posed a new challenge for us. To solve this problem, we introduce a novel Memory (and Time) Efficient Image Tracing (MEIT) framework. Evaluated on the Gold dataset, our proposed method achieves better or competitive performance compared to state-of-the-art neuron tracing methods in most cases while requiring less memory and time.
{"title":"Memory and Time Efficient 3D Neuron Morphology Tracing in Large-Scale Images","authors":"Heng Wang, Donghao Zhang, Yang Song, Siqi Liu, Rong Gao, Hanchuan Peng, Weidong (Tom) Cai","doi":"10.1109/DICTA.2018.8615765","DOIUrl":"https://doi.org/10.1109/DICTA.2018.8615765","url":null,"abstract":"3D reconstruction of neuronal morphology is crucial to solving neuron-related problems in neuroscience as it is a key technique for investigating the connectivity and functionality of the neuron system. Many methods have been proposed to improve the accuracy of digital neuron reconstruction. However, the large amount of computer memory and computation time they require to process the large-scale images have posed a new challenge for us. To solve this problem, we introduce a novel Memory (and Time) Efficient Image Tracing (MEIT) framework. Evaluated on the Gold dataset, our proposed method achieves better or competitive performance compared to state-of-the-art neuron tracing methods in most cases while requiring less memory and time.","PeriodicalId":130057,"journal":{"name":"2018 Digital Image Computing: Techniques and Applications (DICTA)","volume":"100 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115748643","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 : 2018-12-01DOI: 10.1109/DICTA.2018.8615800
Md Tahmid Hossain, S. Teng, Dengsheng Zhang, Suryani Lim, Guojun Lu
Image registration has received great attention from researchers over the last few decades. SIFT (Scale Invariant Feature Transform), a local descriptor-based technique is widely used for registering and matching images. To establish correspondences between images, SIFT uses a Euclidean Distance ratio metric. However, this approach leads to a lot of incorrect matches and eliminating these inaccurate matches has been a challenge. Various methods have been proposed attempting to mitigate this problem. In this paper, we propose a scale and orientation harmony-based pruning method that improves image matching process by successfully eliminating incorrect SIFT descriptor matches. Moreover, our technique can predict the image transformation parameters based on a novel adaptive clustering method with much higher matching accuracy. Our experimental results have shown that the proposed method has achieved averages of approximately 16% and 10% higher matching accuracy compared to the traditional SIFT and a contemporary method respectively.
{"title":"Enhancing the Effectiveness of Local Descriptor Based Image Matching","authors":"Md Tahmid Hossain, S. Teng, Dengsheng Zhang, Suryani Lim, Guojun Lu","doi":"10.1109/DICTA.2018.8615800","DOIUrl":"https://doi.org/10.1109/DICTA.2018.8615800","url":null,"abstract":"Image registration has received great attention from researchers over the last few decades. SIFT (Scale Invariant Feature Transform), a local descriptor-based technique is widely used for registering and matching images. To establish correspondences between images, SIFT uses a Euclidean Distance ratio metric. However, this approach leads to a lot of incorrect matches and eliminating these inaccurate matches has been a challenge. Various methods have been proposed attempting to mitigate this problem. In this paper, we propose a scale and orientation harmony-based pruning method that improves image matching process by successfully eliminating incorrect SIFT descriptor matches. Moreover, our technique can predict the image transformation parameters based on a novel adaptive clustering method with much higher matching accuracy. Our experimental results have shown that the proposed method has achieved averages of approximately 16% and 10% higher matching accuracy compared to the traditional SIFT and a contemporary method respectively.","PeriodicalId":130057,"journal":{"name":"2018 Digital Image Computing: Techniques and Applications (DICTA)","volume":"94 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124258544","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 : 2018-12-01DOI: 10.1109/DICTA.2018.8615844
N. Abid, A. Ul-Hasan, F. Shafait
Postal applications are among the first beneficiaries of the advancements in document image processing techniques due to their economic significance. To automate the process of postal services, it is necessary to integrate contributions from a wide range of image processing domains, from image acquisition and preprocessing to interpretation through symbol, character and word recognition. Lately, machine learning approaches are deployed for postal address processing. Parsing problem has been explored using different techniques, like regular expressions, Conditional Random Fields (CRFs), Hidden Markov Models (HMMs), Decision Trees and Support Vector Machines (SVMs). These traditional techniques are designed on the assumption that the data is free from OCR errors which decreases the adaptability of the architecture in the real-world scenarios. Furthermore, their performance is affected in the presence of non-standardized addresses resulting in intermixing of similar classes. In this paper, we present the first trainable neural network based robust architecture DeepParse for postal address parsing that tackles these issues and can be applied to any Named Entity Recognition (NER) problem. The architecture takes the input at different granularity levels: characters, trigram characters and words to extract and learn the features and classify the addresses. The model was trained on a synthetically generated dataset and tested on the real-world addresses. DeepParse has also been tested on the NER dataset i.e. CoNLL2003 and gave the result of 90.44% which is on par with the state-of-art technique.
{"title":"DeepParse: A Trainable Postal Address Parser","authors":"N. Abid, A. Ul-Hasan, F. Shafait","doi":"10.1109/DICTA.2018.8615844","DOIUrl":"https://doi.org/10.1109/DICTA.2018.8615844","url":null,"abstract":"Postal applications are among the first beneficiaries of the advancements in document image processing techniques due to their economic significance. To automate the process of postal services, it is necessary to integrate contributions from a wide range of image processing domains, from image acquisition and preprocessing to interpretation through symbol, character and word recognition. Lately, machine learning approaches are deployed for postal address processing. Parsing problem has been explored using different techniques, like regular expressions, Conditional Random Fields (CRFs), Hidden Markov Models (HMMs), Decision Trees and Support Vector Machines (SVMs). These traditional techniques are designed on the assumption that the data is free from OCR errors which decreases the adaptability of the architecture in the real-world scenarios. Furthermore, their performance is affected in the presence of non-standardized addresses resulting in intermixing of similar classes. In this paper, we present the first trainable neural network based robust architecture DeepParse for postal address parsing that tackles these issues and can be applied to any Named Entity Recognition (NER) problem. The architecture takes the input at different granularity levels: characters, trigram characters and words to extract and learn the features and classify the addresses. The model was trained on a synthetically generated dataset and tested on the real-world addresses. DeepParse has also been tested on the NER dataset i.e. CoNLL2003 and gave the result of 90.44% which is on par with the state-of-art technique.","PeriodicalId":130057,"journal":{"name":"2018 Digital Image Computing: Techniques and Applications (DICTA)","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124345093","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 : 2018-12-01DOI: 10.1109/DICTA.2018.8615786
S. Kabbour, Pierre-Yves Richard
In this paper, a novel fully automated method is developed to acquire an accurate surface 3D reconstruction of the human ear by using multi-view stereo vision and morphable model without texture. As the results show, our method outperform state of the art approaches. Our method is based on using a template to estimate the pose and orientation of the camera without relying on correspondences, and after dense reconstruction is done, the ear morphable model is fitted on this point cloud by minimizing the distance between them, the form of the model can be transform as wished by its coefficients, and it only uses shape without relying on texture to converge its coefficients.
{"title":"Human Ear Surface Reconstruction Through Morphable Model Deformation","authors":"S. Kabbour, Pierre-Yves Richard","doi":"10.1109/DICTA.2018.8615786","DOIUrl":"https://doi.org/10.1109/DICTA.2018.8615786","url":null,"abstract":"In this paper, a novel fully automated method is developed to acquire an accurate surface 3D reconstruction of the human ear by using multi-view stereo vision and morphable model without texture. As the results show, our method outperform state of the art approaches. Our method is based on using a template to estimate the pose and orientation of the camera without relying on correspondences, and after dense reconstruction is done, the ear morphable model is fitted on this point cloud by minimizing the distance between them, the form of the model can be transform as wished by its coefficients, and it only uses shape without relying on texture to converge its coefficients.","PeriodicalId":130057,"journal":{"name":"2018 Digital Image Computing: Techniques and Applications (DICTA)","volume":"87 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114906367","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 : 2018-12-01DOI: 10.1109/DICTA.2018.8615866
M. Schellhorn, G. Notni
For the acceptance of spectral measurement technology for quality assurance and inspection in the industrial sector, the acquisition and processing of spectral images must be adapted to the production cycle. When processing spectral images, variations of the Principal Component Analysis (PCA) are often used as preprocessing steps, for example for segmentation, spectral decomposition or data compression. To speed up this time-consuming algorithm, hardware and software cores were implemented on a system-on-a-programmable-chip (SoPC). This paper deals with the optimization of this implementation to minimize calculation times. Special attention is paid to the cores used to calculate covariances and data derivation. The restructuring of the hardware IP (Intellectual property) cores and fundamental design decisions are discussed. The optimization was implemented and evaluated on a 12-channel spectral camera.
{"title":"Optimization of a Principal Component Analysis Implementation on Field-Programmable Gate Arrays (FPGA) for Analysis of Spectral Images","authors":"M. Schellhorn, G. Notni","doi":"10.1109/DICTA.2018.8615866","DOIUrl":"https://doi.org/10.1109/DICTA.2018.8615866","url":null,"abstract":"For the acceptance of spectral measurement technology for quality assurance and inspection in the industrial sector, the acquisition and processing of spectral images must be adapted to the production cycle. When processing spectral images, variations of the Principal Component Analysis (PCA) are often used as preprocessing steps, for example for segmentation, spectral decomposition or data compression. To speed up this time-consuming algorithm, hardware and software cores were implemented on a system-on-a-programmable-chip (SoPC). This paper deals with the optimization of this implementation to minimize calculation times. Special attention is paid to the cores used to calculate covariances and data derivation. The restructuring of the hardware IP (Intellectual property) cores and fundamental design decisions are discussed. The optimization was implemented and evaluated on a 12-channel spectral camera.","PeriodicalId":130057,"journal":{"name":"2018 Digital Image Computing: Techniques and Applications (DICTA)","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128897239","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 : 2018-12-01DOI: 10.1109/DICTA.2018.8615798
Yukun Yang, Min Xu, Wanneng Wu, Ruiheng Zhang, Yu Peng
This paper addresses the issue of 3D multiview basketball players detection and localization. Existing methods for this problem typically take background subtraction as input, which limits the accuracy of localization and the performance of further object tracking. Moreover, the performance of background subtraction based methods is heavily impacted by the occlusions in crowded scenes. In this paper, we propose an innovative method which jointly implements deep learning based player detection and occupancy probability based player localization. What's more, a new Bayesian model of the localization algorithms is developed, which uses foreground information from fisheye cameras to setup meaningful initialization values in the first step of iteration, in order to not only eliminate ambiguous detection, but also accelerate computational processes. Experimental results on real basketball game data demonstrate that our methods significantly improve the performance compared with current methods, by eliminating missed and false detection, as well as increasing probabilities of positive results.
{"title":"3D Multiview Basketball Players Detection and Localization Based on Probabilistic Occupancy","authors":"Yukun Yang, Min Xu, Wanneng Wu, Ruiheng Zhang, Yu Peng","doi":"10.1109/DICTA.2018.8615798","DOIUrl":"https://doi.org/10.1109/DICTA.2018.8615798","url":null,"abstract":"This paper addresses the issue of 3D multiview basketball players detection and localization. Existing methods for this problem typically take background subtraction as input, which limits the accuracy of localization and the performance of further object tracking. Moreover, the performance of background subtraction based methods is heavily impacted by the occlusions in crowded scenes. In this paper, we propose an innovative method which jointly implements deep learning based player detection and occupancy probability based player localization. What's more, a new Bayesian model of the localization algorithms is developed, which uses foreground information from fisheye cameras to setup meaningful initialization values in the first step of iteration, in order to not only eliminate ambiguous detection, but also accelerate computational processes. Experimental results on real basketball game data demonstrate that our methods significantly improve the performance compared with current methods, by eliminating missed and false detection, as well as increasing probabilities of positive results.","PeriodicalId":130057,"journal":{"name":"2018 Digital Image Computing: Techniques and Applications (DICTA)","volume":"61 2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130720214","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 : 2018-12-01DOI: 10.1109/DICTA.2018.8615799
Kosaku Fujita, Masayuki Kobayashi, T. Nagao
In the machine learning community, data augmentation techniques have been widely used to make deep neural networks invariant to object transition. However, less attention has been paid to data augmentation in traditional classification methods. In this paper, we take a closer look at traditional classification methods and introduce a new data augmentation technique based on the concept of image transformation. Starting with a few existing examples, we add noise and generate new data points to reduce sparseness in a given feature space. Then, we generate images corresponding to the new data points, although this is usually an ill-posed problem. Herein, the novelty is in constructing an image transformation tree and generating new data from a small number of instances. This allows us to reduce sparseness in the feature space and build more robust classifiers. We evaluate our method on the Caltech-101 dataset to verify its potential. In the context of the situation where the amount of training data is limited, we demonstrate that the support vector machine-based classifiers trained with an augmented dataset using our method outperform classifiers trained with the original dataset in most cases.
{"title":"Data Augmentation using Evolutionary Image Processing","authors":"Kosaku Fujita, Masayuki Kobayashi, T. Nagao","doi":"10.1109/DICTA.2018.8615799","DOIUrl":"https://doi.org/10.1109/DICTA.2018.8615799","url":null,"abstract":"In the machine learning community, data augmentation techniques have been widely used to make deep neural networks invariant to object transition. However, less attention has been paid to data augmentation in traditional classification methods. In this paper, we take a closer look at traditional classification methods and introduce a new data augmentation technique based on the concept of image transformation. Starting with a few existing examples, we add noise and generate new data points to reduce sparseness in a given feature space. Then, we generate images corresponding to the new data points, although this is usually an ill-posed problem. Herein, the novelty is in constructing an image transformation tree and generating new data from a small number of instances. This allows us to reduce sparseness in the feature space and build more robust classifiers. We evaluate our method on the Caltech-101 dataset to verify its potential. In the context of the situation where the amount of training data is limited, we demonstrate that the support vector machine-based classifiers trained with an augmented dataset using our method outperform classifiers trained with the original dataset in most cases.","PeriodicalId":130057,"journal":{"name":"2018 Digital Image Computing: Techniques and Applications (DICTA)","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132999215","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 : 2018-12-01DOI: 10.1109/DICTA.2018.8615805
M. Shoeiby, M. Armin, A. Robles-Kelly
In this paper, we present a method for registration of image pairs. Our method relates both images to one another for registration purposes making use of optical flow. We formulate the problem in a variational setting making use of an L1-norm fidelity term, a total variation (TV) criterion, and a geometric constraint. This treatment leads to a cost function, in which, both the total variation and the homographic constraints are enforced via regularisation. Further, to compute the flow we employ a multiscale pyramid, whereby the total variation is minimized at each layer and the geometric constraint is enforced between layers. In practice, this is carried out by using a Rudin-Osher-Fatemi (ROF) denoising model within each layer and a gated function for the homography computation between layers. We also illustrate the utility of our method for image registration and flow computation and compare our approach to a mainstream non-geometrically constrained variational alternative elsewhere in the literature.
{"title":"Image Registration via Geometrically Constrained Total Variation Optical Flow","authors":"M. Shoeiby, M. Armin, A. Robles-Kelly","doi":"10.1109/DICTA.2018.8615805","DOIUrl":"https://doi.org/10.1109/DICTA.2018.8615805","url":null,"abstract":"In this paper, we present a method for registration of image pairs. Our method relates both images to one another for registration purposes making use of optical flow. We formulate the problem in a variational setting making use of an L1-norm fidelity term, a total variation (TV) criterion, and a geometric constraint. This treatment leads to a cost function, in which, both the total variation and the homographic constraints are enforced via regularisation. Further, to compute the flow we employ a multiscale pyramid, whereby the total variation is minimized at each layer and the geometric constraint is enforced between layers. In practice, this is carried out by using a Rudin-Osher-Fatemi (ROF) denoising model within each layer and a gated function for the homography computation between layers. We also illustrate the utility of our method for image registration and flow computation and compare our approach to a mainstream non-geometrically constrained variational alternative elsewhere in the literature.","PeriodicalId":130057,"journal":{"name":"2018 Digital Image Computing: Techniques and Applications (DICTA)","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122425470","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 : 2018-12-01DOI: 10.1109/DICTA.2018.8615863
Yutaro Katano, Tetsuhiko Muroi, N. Kinoshita, Norihiko Ishii
We evaluated a deep learning-based data demodulation method for multi-level recording data in holographic data storage. This method demodulates reproduced data as pattern recognition using a convolutional neural network. The network learns the rule of demodulation in consideration of optical noise that deteriorates the quality of reproduced data. Unlike with a conventional hard decision method, the learnt network demodulated the noise-added data accurately and decreased demodulation errors.
{"title":"Demodulation of Multi-Level Data using Convolutional Neural Network in Holographic Data Storage","authors":"Yutaro Katano, Tetsuhiko Muroi, N. Kinoshita, Norihiko Ishii","doi":"10.1109/DICTA.2018.8615863","DOIUrl":"https://doi.org/10.1109/DICTA.2018.8615863","url":null,"abstract":"We evaluated a deep learning-based data demodulation method for multi-level recording data in holographic data storage. This method demodulates reproduced data as pattern recognition using a convolutional neural network. The network learns the rule of demodulation in consideration of optical noise that deteriorates the quality of reproduced data. Unlike with a conventional hard decision method, the learnt network demodulated the noise-added data accurately and decreased demodulation errors.","PeriodicalId":130057,"journal":{"name":"2018 Digital Image Computing: Techniques and Applications (DICTA)","volume":"21 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122487442","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}
Palmprint recognition is a very important field of biometrics, and has been intensively researched on both feature extraction and classification methods. Recently, deep learning techniques such as convolutional neural networks have demonstrated clear advantages over traditional learning algorithms for various image classification tasks such as object recognition and detection. However, a large amount of data is needed to train deep networks, which limits its application to some tasks such as palmprint recognition where it lacks of sufficient training samples for each class (i.e., each individual). In this paper, we propose a Generative Adversarial Net (GAN) based solution to augment training data for improved performance of palmprint recognition. An improved Deep Convolutional Generative Adversarial Net (DCGAN) is first devised to generate high quality plamprint images by replacing convolutional transpose layer with linear upsampling and introducing Structure Similarity (SSIM) index into loss function. As a result, the generated images have discriminative features, increased smoothness and consistency, and less variance compared to those generated by the baseline DCGAN. Then, a mixing training strategy via a combination of GAN-based and classical data augmentation techniques is adopted to further improve recognition performance. The experimental results on two publicly available datasets demonstrate the effectiveness of our proposed GAN based data augmentation method in palmprint recognition. Our method is able to achieve 1.52% and 0.37% Equal Error Rates (EER) on IIT Delhi and CASIA palmprint datasets, respectively, which outperforms other existing methods.
{"title":"Generative Adversarial Network (GAN) Based Data Augmentation for Palmprint Recognition","authors":"Gengxing Wang, Wenxiong Kang, Qiuxia Wu, Zhiyong Wang, Junbin Gao","doi":"10.1109/DICTA.2018.8615782","DOIUrl":"https://doi.org/10.1109/DICTA.2018.8615782","url":null,"abstract":"Palmprint recognition is a very important field of biometrics, and has been intensively researched on both feature extraction and classification methods. Recently, deep learning techniques such as convolutional neural networks have demonstrated clear advantages over traditional learning algorithms for various image classification tasks such as object recognition and detection. However, a large amount of data is needed to train deep networks, which limits its application to some tasks such as palmprint recognition where it lacks of sufficient training samples for each class (i.e., each individual). In this paper, we propose a Generative Adversarial Net (GAN) based solution to augment training data for improved performance of palmprint recognition. An improved Deep Convolutional Generative Adversarial Net (DCGAN) is first devised to generate high quality plamprint images by replacing convolutional transpose layer with linear upsampling and introducing Structure Similarity (SSIM) index into loss function. As a result, the generated images have discriminative features, increased smoothness and consistency, and less variance compared to those generated by the baseline DCGAN. Then, a mixing training strategy via a combination of GAN-based and classical data augmentation techniques is adopted to further improve recognition performance. The experimental results on two publicly available datasets demonstrate the effectiveness of our proposed GAN based data augmentation method in palmprint recognition. Our method is able to achieve 1.52% and 0.37% Equal Error Rates (EER) on IIT Delhi and CASIA palmprint datasets, respectively, which outperforms other existing methods.","PeriodicalId":130057,"journal":{"name":"2018 Digital Image Computing: Techniques and Applications (DICTA)","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2018-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122549782","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：