Stuart Gibson | University of Kent (original) (raw)

Papers by Stuart Gibson

arXiv (Cornell University), Jun 23, 2018

Convolutional neural networks (CNN) have been shown to provide a good solution for classification... more Convolutional neural networks (CNN) have been shown to provide a good solution for classification problems that utilize data obtained from vibrational spectroscopy. Moreover, CNNs are capable of identification from noisy spectra without the need for additional preprocessing. However, their application in practical spectroscopy is limited due to two shortcomings. The e ectiveness of the classification using CNNs drops rapidly when only a small number of spectra per substance are available for training (which is a typical situation in real applications). Additionally, to accommodate new, previously unseen substance classes, the network must be retrained which is computationally intensive. Here we address these issues by reformulating a multi-class classification problem with a large number of classes, but a small number of samples per class, to a binary classification problem with su icient data available for representation learning. Namely, we define the learning task as identifying pairs of inputs as belonging to the same or di erent classes. We achieve this using a Siamese convolutional neural network. A novel sampling strategy is proposed to address the imbalance problem in training the Siamese Network. The trained network can e ectively classify samples of unseen substance classes using just a single reference sample (termed as one-shot learning in the machine learning community). Our results demonstrate be er accuracy than other practical systems to date, while allowing e ortless updates of the system's database with novel substance classes.

arXiv (Cornell University), Apr 27, 2020

Recent work showed neural-network based approaches to reconstructing images from compressively se... more Recent work showed neural-network based approaches to reconstructing images from compressively sensed measurements offer significant improvements in accuracy and signal compression. Such methods can dramatically boost the capability of computational imaging hardware. However, to date, there have been two major drawbacks: (1) the high-precision real-valued sensing patterns proposed in the majority of existing works can prove problematic when used with computational imaging hardware such as a digital micromirror sampling device and (2) the network structures for image reconstruction involve intensive computation, which is also not suitable for hardware deployment. To address these problems, we propose a novel hardware-friendly solution based on mixed-weights neural networks for computational imaging. In particular, learned binary-weight sensing patterns are tailored to the sampling device. Moreover, we proposed a recursive network structure for low-resolution image sampling and high-resolution reconstruction scheme. It reduces both the required number of measurements and reconstruction computation by operating convolution on small intermediate feature maps. The recursive structure further reduced the model size, making the network more computationally efficient when deployed with the hardware. Our method has been validated on benchmark datasets and achieved state of the art reconstruction accuracy. We tested our proposed network in conjunction with a proof-of-concept hardware setup.

Analytical Methods, 2013

The version in the Kent Academic Repository may differ from the final published version. Users ar... more The version in the Kent Academic Repository may differ from the final published version. Users are advised to check http://kar.kent.ac.uk for the status of the paper. Users should always cite the published version of record.

arXiv (Cornell University), Mar 8, 2018

An ultrafast single-pixel optical 2D imaging system using a single multimode fiber (MF) is propos... more An ultrafast single-pixel optical 2D imaging system using a single multimode fiber (MF) is proposed. The MF acted as the all-optical random pattern generator. Light with different wavelengths pass through a single MF will generator all-optical random speckle patterns, which have a low correlation of 0.074 with 0.1nm wavelength step from 1518.0nm to 1567.9nm. The all-optical random speckle patterns are perfect for compressive sensing (CS) imaging with the advantage of low cost in comparison with the conventional expensive pseudorandom binary sequence (PRBS). Besides, with the employment of photonic time stretch (PTS), light of different wavelengths will go through a single capsuled MF in time serial within a short pulse time, which makes ultrafast single-pixel all-optical CS imaging possible. In our work, the all-optical random speckle patterns are analyzed and used to perform CS imaging in our proposed system and the results shows a single-pixel photo-detector can be employed in CS imaging system and a 27 × 27 pixels image is reconstructed within 500 measurements. In our proposed imaging system, the fast Fourier transform (FFT) spatial resolution, which is a combination of multiple Gaussians, is analyzed. Considering 4 optical speckle patterns, the FFT spatial resolution is 50 × 50 pixels. This resolution limit has been obtained by removing the central low frequency components and observing the significant spectral power along all the radial directions.

Journal of Physics: Condensed Matter, 2021

In recent years, artificial intelligence techniques have proved to be very successful when applie... more In recent years, artificial intelligence techniques have proved to be very successful when applied to problems in physical sciences. Here we apply an unsupervised machine learning (ML) algorithm called principal component analysis (PCA) as a tool to analyse the data from muon spectroscopy experiments. Specifically, we apply the ML technique to detect phase transitions in various materials. The measured quantity in muon spectroscopy is an asymmetry function, which may hold information about the distribution of the intrinsic magnetic field in combination with the dynamics of the sample. Sharp changes of shape of asymmetry functions—measured at different temperatures—might indicate a phase transition. Existing methods of processing the muon spectroscopy data are based on regression analysis, but choosing the right fitting function requires knowledge about the underlying physics of the probed material. Conversely, PCA focuses on small differences in the asymmetry curves and works withou...

2nd Canterbury Conference on OCT with Emphasis on Broadband Optical Sources, 2018

The version in the Kent Academic Repository may differ from the final published version. Users ar... more The version in the Kent Academic Repository may differ from the final published version. Users are advised to check http://kar.kent.ac.uk for the status of the paper. Users should always cite the published version of record.

ArXiv, 2017

In this paper we present a new method for cystoid macular edema (CME) segmentation in retinal Opt... more In this paper we present a new method for cystoid macular edema (CME) segmentation in retinal Optical Coherence Tomography (OCT) images, using a fully convolutional neural network (FCN) and a fully connected conditional random fields (dense CRFs). As a first step, the framework trains the FCN model to extract features from retinal layers in OCT images, which exhibit CME, and then segments CME regions using the trained model. Thereafter, dense CRFs are used to refine the segmentation according to the edema appearance. We have trained and tested the framework with OCT images from 10 patients with diabetic macular edema (DME). Our experimental results show that fluid and concrete macular edema areas were segmented with good adherence to boundaries. A segmentation accuracy of 0.61pm0.210.61\pm 0.210.61pm0.21 (Dice coefficient) was achieved, with respect to the ground truth, which compares favourably with the previous state-of-the-art that used a kernel regression based method ($0.51\pm 0.34$). Our approa...

Informatica (Slovenia), 2015

Facial composite construction is one of the most successful applications of interactive evolution... more Facial composite construction is one of the most successful applications of interactive evolutionary computation. In spite of this, previous work in the area of composite construction has not investigated the algorithm design options in detail. We address this issue with four experiments. In the first experiment a sorting task is used to identify the 12 most salient dimensions of a 30-dimensional search space. In the second experiment the performances of two mutation and two recombination operators for interactive genetic algorithms are compared. In the third experiment three search spaces are compared: a 30-dimensional search space, a mathematically reduced 12-dimensional search space, and a 12-dimensional search space formed from the 12 most salient dimensions. Finally, we compare the performances of an interactive genetic algorithm to interactive differential evolution. Our results show that the facial composite construction process is remarkably robust to the choice of evolution...

In this paper a series of experiments concerning the use of IEAs in the creation of facial compos... more In this paper a series of experiments concerning the use of IEAs in the creation of facial composites are reported. A human evaluation based search space, which is itself a subspace of a larger search space, is created. The human reduced search space is used to compare two mutation operators and two recombination operators in an IEA. A mathematically reduced search space is constructed from the larger search space. The facial composite process is performed in the three search spaces. No statistically significant differences are found between the performances of the operators or the search spaces.

ArXiv, 2016

One of the main challenges faced by Biometric-based authentication systems is the need to offer s... more One of the main challenges faced by Biometric-based authentication systems is the need to offer secure authentication while maintaining the privacy of the biometric data. Previous solutions, such as Secure Sketch and Fuzzy Extractors, rely on assumptions that cannot be guaranteed in practice, and often affect the authentication accuracy. In this paper, we introduce HoneyFaces: the concept of adding a large set of synthetic faces (indistinguishable from real) into the biometric "password file". This password inflation protects the privacy of users and increases the security of the system without affecting the accuracy of the authentication. In particular, privacy for the real users is provided by "hiding" them among a large number of fake users (as the distributions of synthetic and real faces are equal). In addition to maintaining the authentication accuracy, and thus not affecting the security of the authentication process, HoneyFaces offer several security impr...

ArXiv, 2018

Deep Learning shows very good performance when trained on large labeled data sets. The problem of... more Deep Learning shows very good performance when trained on large labeled data sets. The problem of training a deep net on a few or one sample per class requires a different learning approach which can generalize to unseen classes using only a few representatives of these classes. This problem has previously been approached by meta-learning. Here we propose a novel meta-learner which shows state-of-the-art performance on common benchmarks for one/few shot classification. Our model features three novel components: First is a feed-forward embedding that takes random class support samples (after a customary CNN embedding) and transfers them to a better class representation in terms of a classification problem. Second is a novel attention mechanism, inspired by competitive learning, which causes class representatives to compete with each other to become a temporary class prototype with respect to the query point. This mechanism allows switching between representatives depending on the pos...

IET Image Processing, 2019

The version in the Kent Academic Repository may differ from the final published version. Users ar... more The version in the Kent Academic Repository may differ from the final published version. Users are advised to check http://kar.kent.ac.uk for the status of the paper. Users should always cite the published version of record.

Chemometrics and Intelligent Laboratory Systems, 2018

The version in the Kent Academic Repository may differ from the final published version. Users ar... more The version in the Kent Academic Repository may differ from the final published version. Users are advised to check http://kar.kent.ac.uk for the status of the paper. Users should always cite the published version of record.

Psychophysiology, 2018

Recently, we showed that presenting salient names (i.e., a participant's first name) on the f... more Recently, we showed that presenting salient names (i.e., a participant's first name) on the fringe of awareness (in rapid serial visual presentation, RSVP) breaks through into awareness, resulting in the generation of a P3, which (if concealed information is presented) could be used to differentiate between deceivers and nondeceivers. The aim of the present study was to explore whether face stimuli can be used in an ERP‐based RSVP paradigm to infer recognition of broadly familiar faces. To do this, we explored whether famous faces differentially break into awareness when presented in RSVP and, importantly, whether ERPs can be used to detect these breakthrough events on an individual basis. Our findings provide evidence that famous faces are differentially perceived and processed by participants’ brains as compared to novel (or unfamiliar) faces. EEG data revealed large differences in brain responses between these conditions.

The Analyst, Jan 10, 2017

Machine learning methods have found many applications in Raman spectroscopy, especially for the i... more Machine learning methods have found many applications in Raman spectroscopy, especially for the identification of chemical species. However, almost all of these methods require non-trivial preprocessing such as baseline correction and/or PCA as an essential step. Here we describe our unified solution for the identification of chemical species in which a convolutional neural network is trained to automatically identify substances according to their Raman spectrum without the need for preprocessing. We evaluated our approach using the RRUFF spectral database, comprising mineral sample data. Superior classification performance is demonstrated compared with other frequently used machine learning algorithms including the popular support vector machine method.

IEEE Transactions on Information Forensics and Security, 2017

Recent advances in deep learning (DL) allow for solving complex AI problems that used to be consi... more Recent advances in deep learning (DL) allow for solving complex AI problems that used to be considered very hard. While this progress has advanced many fields, it is considered to be bad news for Completely Automated Public Turing tests to tell Computers and Humans Apart (CAPTCHAs), the security of which rests on the hardness of some learning problems. In this paper, we introduce DeepCAPTCHA, a new and secure CAPTCHA scheme based on adversarial examples, an inherit limitation of the current DL networks. These adversarial examples are constructed inputs, either synthesized from scratch or computed by adding a small and specific perturbation called adversarial noise to correctly classified items, causing the targeted DL network to misclassify them. We show that plain adversarial noise is insufficient to achieve secure CAPTCHA schemes, which leads us to introduce immutable adversarial noise-an adversarial noise that is resistant to removal attempts. In this paper, we implement a proof of concept system, and its analysis shows that the scheme offers high security and good usability compared with the best previously existing CAPTCHAs.

IEEE Photonics Journal, 2017

Photonic time stretch enables real-time high-throughput optical coherence tomography (OCT), but w... more Photonic time stretch enables real-time high-throughput optical coherence tomography (OCT), but with massive data volume being a real challenge. In this paper, data compression in high-throughput optical time-stretch OCT has been explored and experimentally demonstrated. This is made possible by exploiting the spectral sparsity of an encoded optical pulse spectrum using a compressive sensing approach. Both randomization and integration have been implemented in the optical domain avoiding electronic bottleneck. A data compression ratio of 66% has been achieved in high-throughput OCT measurements with 1.51-MHz axial scan rate using greatly reduced data sampling rate of 50 MS/s. Potential to improve compression ratio has been exploited. In addition, using a dual pulse integration method, capability of improving frequency measurement resolution in the proposed system has been demonstrated. A number of optimization algorithms for the reconstruction of the frequency-domain OCT signals have been compared in terms of reconstruction accuracy and efficiency. Our results show that the l 1 magic implementation of the primal-dual interior point method offers the best compromise between accuracy and reconstruction time of the time-stretch OCT signal tested.

The version in the Kent Academic Repository may differ from the final published version. Users ar... more The version in the Kent Academic Repository may differ from the final published version. Users are advised to check http://kar.kent.ac.uk for the status of the paper. Users should always cite the published version of record.

5th International Conference on Imaging for Crime Detection and Prevention (ICDP 2013), 2013

The version in the Kent Academic Repository may differ from the final published version. Users ar... more The version in the Kent Academic Repository may differ from the final published version. Users are advised to check http://kar.kent.ac.uk for the status of the paper. Users should always cite the published version of record.

Analytical Methods, 2013

This study demonstrates that Raman spectroscopy is a valuable tool for discriminating between lip... more This study demonstrates that Raman spectroscopy is a valuable tool for discriminating between lipstick samples under a range of forensically relevant situations. Trace amounts of lipstick smears deposited on textile fibres, cigarette butts and paper tissues were analysed. Differentiation of lipstick smears could be achieved with little or no interference from the underlying medium. Lipstick smears on glass slides, cigarette butts and tissues could also be analysed and identified in situ through evidence bags. Using a range of excitation frequencies (473, 633 and 784 nm) was effective in overcoming problems with fluorescent lipstick samples. The majority of the spectra of deposited lipstick samples remained unchanged over a period of up to two years. In some of the aged lipstick spectra, the (C=C) band at 1655 cm-1 and the (=CH) band at 3011 cm-1 were found to decrease in intensity and disappear over time. The use of chemometrics for the characterisation of large numbers of lipstick spectra was explored. Thirty spectra each from ten different lipsticks were analysed by Principal Components Analysis (PCA) and classified using the K-Nearest Neighbours (KNN) classifier. Up to 98.7% correct classification was achieved. Spectra from trace amounts of lipstick smears deposited on fibres were also analysed and classified using the same technique. 100% correct classification of these samples was achieved.

arXiv (Cornell University), Jun 23, 2018

Convolutional neural networks (CNN) have been shown to provide a good solution for classification... more Convolutional neural networks (CNN) have been shown to provide a good solution for classification problems that utilize data obtained from vibrational spectroscopy. Moreover, CNNs are capable of identification from noisy spectra without the need for additional preprocessing. However, their application in practical spectroscopy is limited due to two shortcomings. The e ectiveness of the classification using CNNs drops rapidly when only a small number of spectra per substance are available for training (which is a typical situation in real applications). Additionally, to accommodate new, previously unseen substance classes, the network must be retrained which is computationally intensive. Here we address these issues by reformulating a multi-class classification problem with a large number of classes, but a small number of samples per class, to a binary classification problem with su icient data available for representation learning. Namely, we define the learning task as identifying pairs of inputs as belonging to the same or di erent classes. We achieve this using a Siamese convolutional neural network. A novel sampling strategy is proposed to address the imbalance problem in training the Siamese Network. The trained network can e ectively classify samples of unseen substance classes using just a single reference sample (termed as one-shot learning in the machine learning community). Our results demonstrate be er accuracy than other practical systems to date, while allowing e ortless updates of the system's database with novel substance classes.

arXiv (Cornell University), Apr 27, 2020

Recent work showed neural-network based approaches to reconstructing images from compressively se... more Recent work showed neural-network based approaches to reconstructing images from compressively sensed measurements offer significant improvements in accuracy and signal compression. Such methods can dramatically boost the capability of computational imaging hardware. However, to date, there have been two major drawbacks: (1) the high-precision real-valued sensing patterns proposed in the majority of existing works can prove problematic when used with computational imaging hardware such as a digital micromirror sampling device and (2) the network structures for image reconstruction involve intensive computation, which is also not suitable for hardware deployment. To address these problems, we propose a novel hardware-friendly solution based on mixed-weights neural networks for computational imaging. In particular, learned binary-weight sensing patterns are tailored to the sampling device. Moreover, we proposed a recursive network structure for low-resolution image sampling and high-resolution reconstruction scheme. It reduces both the required number of measurements and reconstruction computation by operating convolution on small intermediate feature maps. The recursive structure further reduced the model size, making the network more computationally efficient when deployed with the hardware. Our method has been validated on benchmark datasets and achieved state of the art reconstruction accuracy. We tested our proposed network in conjunction with a proof-of-concept hardware setup.

Analytical Methods, 2013

The version in the Kent Academic Repository may differ from the final published version. Users ar... more The version in the Kent Academic Repository may differ from the final published version. Users are advised to check http://kar.kent.ac.uk for the status of the paper. Users should always cite the published version of record.

arXiv (Cornell University), Mar 8, 2018

An ultrafast single-pixel optical 2D imaging system using a single multimode fiber (MF) is propos... more An ultrafast single-pixel optical 2D imaging system using a single multimode fiber (MF) is proposed. The MF acted as the all-optical random pattern generator. Light with different wavelengths pass through a single MF will generator all-optical random speckle patterns, which have a low correlation of 0.074 with 0.1nm wavelength step from 1518.0nm to 1567.9nm. The all-optical random speckle patterns are perfect for compressive sensing (CS) imaging with the advantage of low cost in comparison with the conventional expensive pseudorandom binary sequence (PRBS). Besides, with the employment of photonic time stretch (PTS), light of different wavelengths will go through a single capsuled MF in time serial within a short pulse time, which makes ultrafast single-pixel all-optical CS imaging possible. In our work, the all-optical random speckle patterns are analyzed and used to perform CS imaging in our proposed system and the results shows a single-pixel photo-detector can be employed in CS imaging system and a 27 × 27 pixels image is reconstructed within 500 measurements. In our proposed imaging system, the fast Fourier transform (FFT) spatial resolution, which is a combination of multiple Gaussians, is analyzed. Considering 4 optical speckle patterns, the FFT spatial resolution is 50 × 50 pixels. This resolution limit has been obtained by removing the central low frequency components and observing the significant spectral power along all the radial directions.

Journal of Physics: Condensed Matter, 2021

In recent years, artificial intelligence techniques have proved to be very successful when applie... more In recent years, artificial intelligence techniques have proved to be very successful when applied to problems in physical sciences. Here we apply an unsupervised machine learning (ML) algorithm called principal component analysis (PCA) as a tool to analyse the data from muon spectroscopy experiments. Specifically, we apply the ML technique to detect phase transitions in various materials. The measured quantity in muon spectroscopy is an asymmetry function, which may hold information about the distribution of the intrinsic magnetic field in combination with the dynamics of the sample. Sharp changes of shape of asymmetry functions—measured at different temperatures—might indicate a phase transition. Existing methods of processing the muon spectroscopy data are based on regression analysis, but choosing the right fitting function requires knowledge about the underlying physics of the probed material. Conversely, PCA focuses on small differences in the asymmetry curves and works withou...

2nd Canterbury Conference on OCT with Emphasis on Broadband Optical Sources, 2018

The version in the Kent Academic Repository may differ from the final published version. Users ar... more The version in the Kent Academic Repository may differ from the final published version. Users are advised to check http://kar.kent.ac.uk for the status of the paper. Users should always cite the published version of record.

ArXiv, 2017

In this paper we present a new method for cystoid macular edema (CME) segmentation in retinal Opt... more In this paper we present a new method for cystoid macular edema (CME) segmentation in retinal Optical Coherence Tomography (OCT) images, using a fully convolutional neural network (FCN) and a fully connected conditional random fields (dense CRFs). As a first step, the framework trains the FCN model to extract features from retinal layers in OCT images, which exhibit CME, and then segments CME regions using the trained model. Thereafter, dense CRFs are used to refine the segmentation according to the edema appearance. We have trained and tested the framework with OCT images from 10 patients with diabetic macular edema (DME). Our experimental results show that fluid and concrete macular edema areas were segmented with good adherence to boundaries. A segmentation accuracy of 0.61pm0.210.61\pm 0.210.61pm0.21 (Dice coefficient) was achieved, with respect to the ground truth, which compares favourably with the previous state-of-the-art that used a kernel regression based method ($0.51\pm 0.34$). Our approa...

Informatica (Slovenia), 2015

Facial composite construction is one of the most successful applications of interactive evolution... more Facial composite construction is one of the most successful applications of interactive evolutionary computation. In spite of this, previous work in the area of composite construction has not investigated the algorithm design options in detail. We address this issue with four experiments. In the first experiment a sorting task is used to identify the 12 most salient dimensions of a 30-dimensional search space. In the second experiment the performances of two mutation and two recombination operators for interactive genetic algorithms are compared. In the third experiment three search spaces are compared: a 30-dimensional search space, a mathematically reduced 12-dimensional search space, and a 12-dimensional search space formed from the 12 most salient dimensions. Finally, we compare the performances of an interactive genetic algorithm to interactive differential evolution. Our results show that the facial composite construction process is remarkably robust to the choice of evolution...

In this paper a series of experiments concerning the use of IEAs in the creation of facial compos... more In this paper a series of experiments concerning the use of IEAs in the creation of facial composites are reported. A human evaluation based search space, which is itself a subspace of a larger search space, is created. The human reduced search space is used to compare two mutation operators and two recombination operators in an IEA. A mathematically reduced search space is constructed from the larger search space. The facial composite process is performed in the three search spaces. No statistically significant differences are found between the performances of the operators or the search spaces.

ArXiv, 2016

One of the main challenges faced by Biometric-based authentication systems is the need to offer s... more One of the main challenges faced by Biometric-based authentication systems is the need to offer secure authentication while maintaining the privacy of the biometric data. Previous solutions, such as Secure Sketch and Fuzzy Extractors, rely on assumptions that cannot be guaranteed in practice, and often affect the authentication accuracy. In this paper, we introduce HoneyFaces: the concept of adding a large set of synthetic faces (indistinguishable from real) into the biometric "password file". This password inflation protects the privacy of users and increases the security of the system without affecting the accuracy of the authentication. In particular, privacy for the real users is provided by "hiding" them among a large number of fake users (as the distributions of synthetic and real faces are equal). In addition to maintaining the authentication accuracy, and thus not affecting the security of the authentication process, HoneyFaces offer several security impr...

ArXiv, 2018

Deep Learning shows very good performance when trained on large labeled data sets. The problem of... more Deep Learning shows very good performance when trained on large labeled data sets. The problem of training a deep net on a few or one sample per class requires a different learning approach which can generalize to unseen classes using only a few representatives of these classes. This problem has previously been approached by meta-learning. Here we propose a novel meta-learner which shows state-of-the-art performance on common benchmarks for one/few shot classification. Our model features three novel components: First is a feed-forward embedding that takes random class support samples (after a customary CNN embedding) and transfers them to a better class representation in terms of a classification problem. Second is a novel attention mechanism, inspired by competitive learning, which causes class representatives to compete with each other to become a temporary class prototype with respect to the query point. This mechanism allows switching between representatives depending on the pos...

IET Image Processing, 2019

The version in the Kent Academic Repository may differ from the final published version. Users ar... more The version in the Kent Academic Repository may differ from the final published version. Users are advised to check http://kar.kent.ac.uk for the status of the paper. Users should always cite the published version of record.

Chemometrics and Intelligent Laboratory Systems, 2018

The version in the Kent Academic Repository may differ from the final published version. Users ar... more The version in the Kent Academic Repository may differ from the final published version. Users are advised to check http://kar.kent.ac.uk for the status of the paper. Users should always cite the published version of record.

Psychophysiology, 2018

Recently, we showed that presenting salient names (i.e., a participant's first name) on the f... more Recently, we showed that presenting salient names (i.e., a participant's first name) on the fringe of awareness (in rapid serial visual presentation, RSVP) breaks through into awareness, resulting in the generation of a P3, which (if concealed information is presented) could be used to differentiate between deceivers and nondeceivers. The aim of the present study was to explore whether face stimuli can be used in an ERP‐based RSVP paradigm to infer recognition of broadly familiar faces. To do this, we explored whether famous faces differentially break into awareness when presented in RSVP and, importantly, whether ERPs can be used to detect these breakthrough events on an individual basis. Our findings provide evidence that famous faces are differentially perceived and processed by participants’ brains as compared to novel (or unfamiliar) faces. EEG data revealed large differences in brain responses between these conditions.

The Analyst, Jan 10, 2017

Machine learning methods have found many applications in Raman spectroscopy, especially for the i... more Machine learning methods have found many applications in Raman spectroscopy, especially for the identification of chemical species. However, almost all of these methods require non-trivial preprocessing such as baseline correction and/or PCA as an essential step. Here we describe our unified solution for the identification of chemical species in which a convolutional neural network is trained to automatically identify substances according to their Raman spectrum without the need for preprocessing. We evaluated our approach using the RRUFF spectral database, comprising mineral sample data. Superior classification performance is demonstrated compared with other frequently used machine learning algorithms including the popular support vector machine method.

IEEE Transactions on Information Forensics and Security, 2017

Recent advances in deep learning (DL) allow for solving complex AI problems that used to be consi... more Recent advances in deep learning (DL) allow for solving complex AI problems that used to be considered very hard. While this progress has advanced many fields, it is considered to be bad news for Completely Automated Public Turing tests to tell Computers and Humans Apart (CAPTCHAs), the security of which rests on the hardness of some learning problems. In this paper, we introduce DeepCAPTCHA, a new and secure CAPTCHA scheme based on adversarial examples, an inherit limitation of the current DL networks. These adversarial examples are constructed inputs, either synthesized from scratch or computed by adding a small and specific perturbation called adversarial noise to correctly classified items, causing the targeted DL network to misclassify them. We show that plain adversarial noise is insufficient to achieve secure CAPTCHA schemes, which leads us to introduce immutable adversarial noise-an adversarial noise that is resistant to removal attempts. In this paper, we implement a proof of concept system, and its analysis shows that the scheme offers high security and good usability compared with the best previously existing CAPTCHAs.

IEEE Photonics Journal, 2017

Photonic time stretch enables real-time high-throughput optical coherence tomography (OCT), but w... more Photonic time stretch enables real-time high-throughput optical coherence tomography (OCT), but with massive data volume being a real challenge. In this paper, data compression in high-throughput optical time-stretch OCT has been explored and experimentally demonstrated. This is made possible by exploiting the spectral sparsity of an encoded optical pulse spectrum using a compressive sensing approach. Both randomization and integration have been implemented in the optical domain avoiding electronic bottleneck. A data compression ratio of 66% has been achieved in high-throughput OCT measurements with 1.51-MHz axial scan rate using greatly reduced data sampling rate of 50 MS/s. Potential to improve compression ratio has been exploited. In addition, using a dual pulse integration method, capability of improving frequency measurement resolution in the proposed system has been demonstrated. A number of optimization algorithms for the reconstruction of the frequency-domain OCT signals have been compared in terms of reconstruction accuracy and efficiency. Our results show that the l 1 magic implementation of the primal-dual interior point method offers the best compromise between accuracy and reconstruction time of the time-stretch OCT signal tested.

The version in the Kent Academic Repository may differ from the final published version. Users ar... more The version in the Kent Academic Repository may differ from the final published version. Users are advised to check http://kar.kent.ac.uk for the status of the paper. Users should always cite the published version of record.

5th International Conference on Imaging for Crime Detection and Prevention (ICDP 2013), 2013

The version in the Kent Academic Repository may differ from the final published version. Users ar... more The version in the Kent Academic Repository may differ from the final published version. Users are advised to check http://kar.kent.ac.uk for the status of the paper. Users should always cite the published version of record.

Analytical Methods, 2013

This study demonstrates that Raman spectroscopy is a valuable tool for discriminating between lip... more This study demonstrates that Raman spectroscopy is a valuable tool for discriminating between lipstick samples under a range of forensically relevant situations. Trace amounts of lipstick smears deposited on textile fibres, cigarette butts and paper tissues were analysed. Differentiation of lipstick smears could be achieved with little or no interference from the underlying medium. Lipstick smears on glass slides, cigarette butts and tissues could also be analysed and identified in situ through evidence bags. Using a range of excitation frequencies (473, 633 and 784 nm) was effective in overcoming problems with fluorescent lipstick samples. The majority of the spectra of deposited lipstick samples remained unchanged over a period of up to two years. In some of the aged lipstick spectra, the (C=C) band at 1655 cm-1 and the (=CH) band at 3011 cm-1 were found to decrease in intensity and disappear over time. The use of chemometrics for the characterisation of large numbers of lipstick spectra was explored. Thirty spectra each from ten different lipsticks were analysed by Principal Components Analysis (PCA) and classified using the K-Nearest Neighbours (KNN) classifier. Up to 98.7% correct classification was achieved. Spectra from trace amounts of lipstick smears deposited on fibres were also analysed and classified using the same technique. 100% correct classification of these samples was achieved.