Guangyi Chen - Academia.edu (original) (raw)

Papers by Guangyi Chen

Image Analysis & Stereology

Patch-based image denoising is a popular topic in recent years. In existing works, the distance b... more Patch-based image denoising is a popular topic in recent years. In existing works, the distance between two patches was calculated as their Euclidian distance. When the noise level is high, this approach may not be desirable in image denoising. In this paper, we propose to extract noise-robust feature vectors from image patches and match the image patches by their Euclidian distance of the feature vectors for grey scale image denoising. Our modification takes advantage of the fact that the mean of the Gaussian white noise is zero. For every patch in the noisy image, we use lines to divide the patch into two regions with equal area and we take the mean of the right region for each line. Hence, a number of features can be extracted. We use these extracted features to match the patches in the noisy image. By introducing feature-based patch matching, our method performs favourably for highly noisy images.

Intelligent Computing Theories and Application, 2016

We propose a novel modification to patch matching in block matching and 3D filtering (BM3D), whic... more We propose a novel modification to patch matching in block matching and 3D filtering (BM3D), which is the state-of-the-art in image denoising. The BM3D calculates the distance between two patches by taking the sum of square of the pixel difference. However, when the noise level is very high, this patch matching technique will be less effective. It is well known that Radon transform is very good at suppressing Gaussian white noise and hence in this paper we use it to extract robust features from the two patches for patch matching in BM3D. Experimental results confirm the effectiveness of our proposed modification to BM3D for image denoising in heavily noisy scenarios. Keywords: Image denoising Á Block matching and 3D filtering (BM3D) Á Radon transform Á Gaussian white noise

Intelligent Computing Theories and Application, 2020

Face recognition under varying lighting conditions is an important topic in many real-life applic... more Face recognition under varying lighting conditions is an important topic in many real-life applications. In this paper, we propose a novel algorithm for illumination invariant face recognition. We first convert the face images to the logarithm domain, which makes the dark regions brighter. We then use contourlet transform to generate face images that are approximately invariant to illumination change and use collaborative representation-based classifier (CRC) to classify the unknown faces to one known class. We set the approximation subband and a few highest frequency contourlet coefficient subbands to zero values, and then perform the inverse contourlet transform to generate illumination invariant face images. Experimental results show that our proposed algorithm outperforms two existing methods for the Extended Yale Face Database B for high noise levels. Nevertheless, our new method is not as good as existing methods for low noise levels. In addition, our new method is comparable to existing methods for the CMU-PIE face database.

Lecture Notes in Computer Science, 2005

In this paper, we present new wavelet shrinkage methods for image denoising. The methods take adv... more In this paper, we present new wavelet shrinkage methods for image denoising. The methods take advantage of the higher order statistical coupling between neighboring wavelet coefficients and their corresponding coefficients in the parent level. We also investigate a multiplying factor for the universal threshold in order to obtain better denoising results. An empirical study of this factor shows that its optimum value is approximately the same for different kinds and sizes of images. Experimental results show that our methods give comparatively higher peak signal to noise ratio (PSNR), require less computation time and produce less visual artifacts compared to other methods.

Image and Vision Computing, 2007

A novel descriptor for pattern recognition is proposed by using dual-tree complex wavelet feature... more A novel descriptor for pattern recognition is proposed by using dual-tree complex wavelet features and SVM. The approximate shiftinvariant property of the dual-tree complex wavelet and its good directional selectivity in 2D make it a very appealing choice for pattern recognition. Recently, SVM has been shown to be very successful in pattern recognition. By combining these two tools we find that better recognition results are obtained. We achieve the highest rates when we use the dual-tree complex wavelet features with the Gaussian radial basis function kernel and the wavelet kernel for recognizing similar handwritten numerals, and when we use the Gaussian radial basis function for palmprint classification. Our findings are that the dual-tree complex wavelets are always better than the scalar wavelet for pattern recognition when SVM is used. Also, among many frequently used SVM kernels, the Gaussian radial basis function kernel and the wavelet kernel are the best for pattern recognition applications.

IEEE Signal Processing Letters, 2013

In this letter, we study the Ramanujan Sums (RS) transform by means of matrix multiplication. The... more In this letter, we study the Ramanujan Sums (RS) transform by means of matrix multiplication. The RS are orthogonal in nature and therefore offer excellent energy conservation capability. The 1-D and 2-D forward RS transforms are easy to calculate, but their inverse transforms are not defined in the literature for non-even function. We solved this problem by using matrix multiplication in this letter. Index Terms-Fourier transform (FT), Gaussian white noise, Ramanujan Sums (RS).

International Journal of Wavelets, Multiresolution and Information Processing, 2014

Ramanujan Sums (RS) have been found to be very successful in signal processing recently. However,... more Ramanujan Sums (RS) have been found to be very successful in signal processing recently. However, as far as we know, the RS have not been applied to image analysis. In this paper, we propose two novel algorithms for image analysis, including moment invariants and pattern recognition. Our algorithms are invariant to the translation, rotation and scaling of the 2D shapes. The RS are robust to Gaussian white noise and occlusion as well. Our algorithms compare favourably to the dual-tree complex wavelet (DTCWT) moments and the Zernike's moments in terms of correct classification rates for three well-known shape datasets.

International Conference on Information Science and Technology, 2011

ABSTRACT

2013 Fourth International Conference on Intelligent Control and Information Processing (ICICIP), 2013

ABSTRACT The denoising of Gaussian additive white noise is a classical problem in signal and imag... more ABSTRACT The denoising of Gaussian additive white noise is a classical problem in signal and image processing. In this paper, we classify the most important wavelet denoising methods into different categories and give a brief overview of each method classified. In general, the recently developed block matching and 3D filtering (BM3D) algorithm performs much better than other existing methods published in the literature. We recommend using this method for image denoising because it is currently one of the state-of-the-art denoising methods. The non-local means method and the optimal spatial adaptation (OSA) method are also very successful methods in image denoising.

2006 Canadian Conference on Electrical and Computer Engineering, 2006

A novel descriptor for invariant pattern recognition is proposed by using the Radon transform, th... more A novel descriptor for invariant pattern recognition is proposed by using the Radon transform, the shift-invariant wavelet packet transform, and the Fourier transform. Experimental results show that the proposed descriptor achieves high recognition rates under different rotation angles and noise levels. It outperforms a previously developed method under the noisy environment

Lecture Notes in Computer Science, 2013

ABSTRACT In this paper, we perform sparse signal analysis by using the Ramanujan Sums (RS). The R... more ABSTRACT In this paper, we perform sparse signal analysis by using the Ramanujan Sums (RS). The RS are orthogonal in nature and therefore offer excellent energy conservation. Our analysis shows that the RS can compress the energy of a periodic impulse chain signal into fewer number of RS coefficients than the Fourier transform (FT). In addition, the RS are faster than the FT in computation time because we can calculate the RS basis functions only once and save them to a file. We can retrieve these RS basis functions for our calculation instead of computing them online. To process a signal of 128 samples, we spend 1.0 millisecond for the RS and 5.82 milliseconds for the FT by using our unoptimized Matlab code.

2013 International Conference on Wavelet Analysis and Pattern Recognition, 2013

ABSTRACT The wavelet transform is a very useful tool for a number of real-life applications. This... more ABSTRACT The wavelet transform is a very useful tool for a number of real-life applications. This is due to its multiresolution representation of signals and its localized time-frequency property. The Ramanujan sums (RS) were introduced to signal processing recently. The RS are orthogonal in nature and therefore offer excellent energy conservation. The RS operate on integers and hence can obtain a reduced quantization error implementation. In this paper, we combine the wavelet transform with the RS transform in order to create a new representation of signals. We are trying to combine the merits of the both transforms and at the same time overcome their shortcomings. Our proposed transform contains much richer features than the wavelet transform, so it could be useful for such applications as time-frequency analysis, pattern recognition and image analysis.

Lecture Notes in Computer Science, 2010

A new palmprint classification method is proposed in this paper by using the wavelet features and... more A new palmprint classification method is proposed in this paper by using the wavelet features and AdaBoost. The method outperforms all other classification methods for the PolyU palmprint database. The novelty of the method is two-fold. On one hand, the combination of wavelet features with AdaBoost has never been proposed for palmprint classification before. On the other hand, a recently

Lecture Notes in Computer Science, 2013

In this paper, an invariant algorithm for object recognition is proposed by using the Radon and F... more In this paper, an invariant algorithm for object recognition is proposed by using the Radon and Fourier transforms. It has been shown that this algorithm is invariant to the translation and rotation of pattern images. The scaling invariance can be achieved by the standard normalization techniques. Our algorithm works even when the center of the pattern object is not aligned well. This advantage is because the Fourier spectra are invariant to spatial shift in the radial direction whereas existing methods assume the centroids are aligned exactly. Experimental results show that the proposed method is better than the Zernike's moments, the dual-tree complex wavelet (DTCWT) moments, and the auto-correlation wavelet moments for one aircraft database and one shape database.

Pattern Recognition, 2003

In this paper, we develop a handwritten numeral recognition descriptor using multiwavelets and ne... more In this paper, we develop a handwritten numeral recognition descriptor using multiwavelets and neural networks. We first trace the contour of the numeral, then normalize and resample the contour so that it is translation- and scale-invariant. We then perform multiwavelet orthonormal shell expansion on the contour to get several resolution levels and the average. Finally, we use the shell coefficients as features to input into a feed-forward neural network to recognize the handwritten numerals. The main advantage of the orthonormal shell decomposition is that it decomposes a signal into multiresolution levels, but without down-sampling. Wavelet transforms with down-sampling can give very different coefficients when the input signal is shifted. This is the main limitation of wavelet transforms in pattern recognition. For the shell expansion, we prefer multiwavelets to scalar wavelets because we have two coordinates x and y for each point on the contour. If we extract features from x and y separately, just as Wunsch et al. did (Pattern Recognition 28 (1995) 1237), then we may not get the best features. In addition, we know that multiwavelets have advantages over scalar wavelets, such as short support, orthogonality, symmetry and higher order of vanishing moments. These properties allow multiwavelets to outperform scalar wavelets in some applications, e.g. signal denoising (IEEE Trans. Signal Process. 46 (12) (1998) 3414). We conducted experiments and found that it is feasible to use multiwavelet features in handwritten numeral recognition.

2012 11th International Conference on Information Science, Signal Processing and their Applications (ISSPA), 2012

In recent years, several metrics have been developed for measuring image visual quality, includin... more In recent years, several metrics have been developed for measuring image visual quality, including the MSSIM and the visual information fidelity (VIF). However, these metrics are not robust to spatial shifts, meaning that when the reference and distorted images are misaligned by a few pixels, these metrics will produce very low scores, which is undesirable. In this paper, we extend the SSIM metric to make it robust to spatial shifts by first pre-processing the input images with the Fast Fourier transform (FFT). We then apply the magnitude of the transformed Fourier coefficients to the existing metrics because these coefficients are shiftinvariant. Our assumption is that if we shift the image by a small amount of pixels, then it will not affect the perceived quality. Experimental results show that the proposed method is attractive for measuring the visual quality of 2D images as it is far less complex than the current approach, which consists in performing global motion estimation to align the input images prior to applying the metrics, and offers better accuracy.

International Journal of Pattern Recognition and Artificial Intelligence, 2011

The denoising of a natural signal/image corrupted by Gaussian white noise is a classical problem ... more The denoising of a natural signal/image corrupted by Gaussian white noise is a classical problem in signal/image processing. However, it is still in its infancy to denoise high dimensional data. In this paper, we extended Sendur and Selesnick's bivariate wavelet thresholding from two-dimensional (2D) image denoising to three-dimensional (3D) data cube denoising. Our study shows that bivariate wavelet thresholding is still valid for 3D data cubes. Experimental results show that bivariate wavelet thresholding on 3D data cube is better than performing 2D bivariate wavelet thresholding on every spectral band separately, VisuShrink, and Chen and Zhu's 3-scale denoising.

2013 International Conference on High Performance Computing & Simulation (HPCS), 2013

Support vector machine (SVM) is one of the most popular classification techniques in pattern reco... more Support vector machine (SVM) is one of the most popular classification techniques in pattern recognition community. However, because of outliers in the training samples, SVM tends to perform poorly under such circumstances. In this paper, we borrow the idea from compressive sensing by introducing an extra term to the objective function of the standard SVM in order to achieve a sparse representation. Furthermore, instead of using the l 0 norm, we adopt the l 1 norm in our sparse SVM. In most cases, our method achieves higher classification rates than the standard SVM because of sparser support vectors and is more robust to outliers in the datasets. Experimental results show that our proposed SVM is efficient in pattern recognition applications.

Pattern Recognition and Image Analysis, 2013

ABSTRACT

Pattern Recognition, 2005

Image denoising by means of wavelet transforms has been an active research topic for many years. ... more Image denoising by means of wavelet transforms has been an active research topic for many years. For a given noisy image, which kind of wavelet and what threshold we use should have significant impact on the quality of the denoised image. In this paper, we use Simulated Annealing to find the customized wavelet filters and the customized threshold corresponding to the given noisy image at the same time. Also, we propose to consider a small neighbourhood around the customized wavelet coefficient to be thresholded for image denoising. Experimental results show that our approach is better than VisuShrink, our NeighShrink with fixed wavelet, and the wiener2 filter that is available in Matlab Image Processing Toolbox. In addition, our NeighShrink with fixed wavelet already outperforms VisuShrink for all the experiments.

Image Analysis & Stereology

Patch-based image denoising is a popular topic in recent years. In existing works, the distance b... more Patch-based image denoising is a popular topic in recent years. In existing works, the distance between two patches was calculated as their Euclidian distance. When the noise level is high, this approach may not be desirable in image denoising. In this paper, we propose to extract noise-robust feature vectors from image patches and match the image patches by their Euclidian distance of the feature vectors for grey scale image denoising. Our modification takes advantage of the fact that the mean of the Gaussian white noise is zero. For every patch in the noisy image, we use lines to divide the patch into two regions with equal area and we take the mean of the right region for each line. Hence, a number of features can be extracted. We use these extracted features to match the patches in the noisy image. By introducing feature-based patch matching, our method performs favourably for highly noisy images.

Intelligent Computing Theories and Application, 2016

We propose a novel modification to patch matching in block matching and 3D filtering (BM3D), whic... more We propose a novel modification to patch matching in block matching and 3D filtering (BM3D), which is the state-of-the-art in image denoising. The BM3D calculates the distance between two patches by taking the sum of square of the pixel difference. However, when the noise level is very high, this patch matching technique will be less effective. It is well known that Radon transform is very good at suppressing Gaussian white noise and hence in this paper we use it to extract robust features from the two patches for patch matching in BM3D. Experimental results confirm the effectiveness of our proposed modification to BM3D for image denoising in heavily noisy scenarios. Keywords: Image denoising Á Block matching and 3D filtering (BM3D) Á Radon transform Á Gaussian white noise

Intelligent Computing Theories and Application, 2020

Face recognition under varying lighting conditions is an important topic in many real-life applic... more Face recognition under varying lighting conditions is an important topic in many real-life applications. In this paper, we propose a novel algorithm for illumination invariant face recognition. We first convert the face images to the logarithm domain, which makes the dark regions brighter. We then use contourlet transform to generate face images that are approximately invariant to illumination change and use collaborative representation-based classifier (CRC) to classify the unknown faces to one known class. We set the approximation subband and a few highest frequency contourlet coefficient subbands to zero values, and then perform the inverse contourlet transform to generate illumination invariant face images. Experimental results show that our proposed algorithm outperforms two existing methods for the Extended Yale Face Database B for high noise levels. Nevertheless, our new method is not as good as existing methods for low noise levels. In addition, our new method is comparable to existing methods for the CMU-PIE face database.

Lecture Notes in Computer Science, 2005

In this paper, we present new wavelet shrinkage methods for image denoising. The methods take adv... more In this paper, we present new wavelet shrinkage methods for image denoising. The methods take advantage of the higher order statistical coupling between neighboring wavelet coefficients and their corresponding coefficients in the parent level. We also investigate a multiplying factor for the universal threshold in order to obtain better denoising results. An empirical study of this factor shows that its optimum value is approximately the same for different kinds and sizes of images. Experimental results show that our methods give comparatively higher peak signal to noise ratio (PSNR), require less computation time and produce less visual artifacts compared to other methods.

Image and Vision Computing, 2007

A novel descriptor for pattern recognition is proposed by using dual-tree complex wavelet feature... more A novel descriptor for pattern recognition is proposed by using dual-tree complex wavelet features and SVM. The approximate shiftinvariant property of the dual-tree complex wavelet and its good directional selectivity in 2D make it a very appealing choice for pattern recognition. Recently, SVM has been shown to be very successful in pattern recognition. By combining these two tools we find that better recognition results are obtained. We achieve the highest rates when we use the dual-tree complex wavelet features with the Gaussian radial basis function kernel and the wavelet kernel for recognizing similar handwritten numerals, and when we use the Gaussian radial basis function for palmprint classification. Our findings are that the dual-tree complex wavelets are always better than the scalar wavelet for pattern recognition when SVM is used. Also, among many frequently used SVM kernels, the Gaussian radial basis function kernel and the wavelet kernel are the best for pattern recognition applications.

IEEE Signal Processing Letters, 2013

In this letter, we study the Ramanujan Sums (RS) transform by means of matrix multiplication. The... more In this letter, we study the Ramanujan Sums (RS) transform by means of matrix multiplication. The RS are orthogonal in nature and therefore offer excellent energy conservation capability. The 1-D and 2-D forward RS transforms are easy to calculate, but their inverse transforms are not defined in the literature for non-even function. We solved this problem by using matrix multiplication in this letter. Index Terms-Fourier transform (FT), Gaussian white noise, Ramanujan Sums (RS).

International Journal of Wavelets, Multiresolution and Information Processing, 2014

Ramanujan Sums (RS) have been found to be very successful in signal processing recently. However,... more Ramanujan Sums (RS) have been found to be very successful in signal processing recently. However, as far as we know, the RS have not been applied to image analysis. In this paper, we propose two novel algorithms for image analysis, including moment invariants and pattern recognition. Our algorithms are invariant to the translation, rotation and scaling of the 2D shapes. The RS are robust to Gaussian white noise and occlusion as well. Our algorithms compare favourably to the dual-tree complex wavelet (DTCWT) moments and the Zernike's moments in terms of correct classification rates for three well-known shape datasets.

International Conference on Information Science and Technology, 2011

ABSTRACT

2013 Fourth International Conference on Intelligent Control and Information Processing (ICICIP), 2013

ABSTRACT The denoising of Gaussian additive white noise is a classical problem in signal and imag... more ABSTRACT The denoising of Gaussian additive white noise is a classical problem in signal and image processing. In this paper, we classify the most important wavelet denoising methods into different categories and give a brief overview of each method classified. In general, the recently developed block matching and 3D filtering (BM3D) algorithm performs much better than other existing methods published in the literature. We recommend using this method for image denoising because it is currently one of the state-of-the-art denoising methods. The non-local means method and the optimal spatial adaptation (OSA) method are also very successful methods in image denoising.

2006 Canadian Conference on Electrical and Computer Engineering, 2006

A novel descriptor for invariant pattern recognition is proposed by using the Radon transform, th... more A novel descriptor for invariant pattern recognition is proposed by using the Radon transform, the shift-invariant wavelet packet transform, and the Fourier transform. Experimental results show that the proposed descriptor achieves high recognition rates under different rotation angles and noise levels. It outperforms a previously developed method under the noisy environment

Lecture Notes in Computer Science, 2013

ABSTRACT In this paper, we perform sparse signal analysis by using the Ramanujan Sums (RS). The R... more ABSTRACT In this paper, we perform sparse signal analysis by using the Ramanujan Sums (RS). The RS are orthogonal in nature and therefore offer excellent energy conservation. Our analysis shows that the RS can compress the energy of a periodic impulse chain signal into fewer number of RS coefficients than the Fourier transform (FT). In addition, the RS are faster than the FT in computation time because we can calculate the RS basis functions only once and save them to a file. We can retrieve these RS basis functions for our calculation instead of computing them online. To process a signal of 128 samples, we spend 1.0 millisecond for the RS and 5.82 milliseconds for the FT by using our unoptimized Matlab code.

2013 International Conference on Wavelet Analysis and Pattern Recognition, 2013

ABSTRACT The wavelet transform is a very useful tool for a number of real-life applications. This... more ABSTRACT The wavelet transform is a very useful tool for a number of real-life applications. This is due to its multiresolution representation of signals and its localized time-frequency property. The Ramanujan sums (RS) were introduced to signal processing recently. The RS are orthogonal in nature and therefore offer excellent energy conservation. The RS operate on integers and hence can obtain a reduced quantization error implementation. In this paper, we combine the wavelet transform with the RS transform in order to create a new representation of signals. We are trying to combine the merits of the both transforms and at the same time overcome their shortcomings. Our proposed transform contains much richer features than the wavelet transform, so it could be useful for such applications as time-frequency analysis, pattern recognition and image analysis.

Lecture Notes in Computer Science, 2010

A new palmprint classification method is proposed in this paper by using the wavelet features and... more A new palmprint classification method is proposed in this paper by using the wavelet features and AdaBoost. The method outperforms all other classification methods for the PolyU palmprint database. The novelty of the method is two-fold. On one hand, the combination of wavelet features with AdaBoost has never been proposed for palmprint classification before. On the other hand, a recently

Lecture Notes in Computer Science, 2013

In this paper, an invariant algorithm for object recognition is proposed by using the Radon and F... more In this paper, an invariant algorithm for object recognition is proposed by using the Radon and Fourier transforms. It has been shown that this algorithm is invariant to the translation and rotation of pattern images. The scaling invariance can be achieved by the standard normalization techniques. Our algorithm works even when the center of the pattern object is not aligned well. This advantage is because the Fourier spectra are invariant to spatial shift in the radial direction whereas existing methods assume the centroids are aligned exactly. Experimental results show that the proposed method is better than the Zernike's moments, the dual-tree complex wavelet (DTCWT) moments, and the auto-correlation wavelet moments for one aircraft database and one shape database.

Pattern Recognition, 2003

In this paper, we develop a handwritten numeral recognition descriptor using multiwavelets and ne... more In this paper, we develop a handwritten numeral recognition descriptor using multiwavelets and neural networks. We first trace the contour of the numeral, then normalize and resample the contour so that it is translation- and scale-invariant. We then perform multiwavelet orthonormal shell expansion on the contour to get several resolution levels and the average. Finally, we use the shell coefficients as features to input into a feed-forward neural network to recognize the handwritten numerals. The main advantage of the orthonormal shell decomposition is that it decomposes a signal into multiresolution levels, but without down-sampling. Wavelet transforms with down-sampling can give very different coefficients when the input signal is shifted. This is the main limitation of wavelet transforms in pattern recognition. For the shell expansion, we prefer multiwavelets to scalar wavelets because we have two coordinates x and y for each point on the contour. If we extract features from x and y separately, just as Wunsch et al. did (Pattern Recognition 28 (1995) 1237), then we may not get the best features. In addition, we know that multiwavelets have advantages over scalar wavelets, such as short support, orthogonality, symmetry and higher order of vanishing moments. These properties allow multiwavelets to outperform scalar wavelets in some applications, e.g. signal denoising (IEEE Trans. Signal Process. 46 (12) (1998) 3414). We conducted experiments and found that it is feasible to use multiwavelet features in handwritten numeral recognition.

2012 11th International Conference on Information Science, Signal Processing and their Applications (ISSPA), 2012

In recent years, several metrics have been developed for measuring image visual quality, includin... more In recent years, several metrics have been developed for measuring image visual quality, including the MSSIM and the visual information fidelity (VIF). However, these metrics are not robust to spatial shifts, meaning that when the reference and distorted images are misaligned by a few pixels, these metrics will produce very low scores, which is undesirable. In this paper, we extend the SSIM metric to make it robust to spatial shifts by first pre-processing the input images with the Fast Fourier transform (FFT). We then apply the magnitude of the transformed Fourier coefficients to the existing metrics because these coefficients are shiftinvariant. Our assumption is that if we shift the image by a small amount of pixels, then it will not affect the perceived quality. Experimental results show that the proposed method is attractive for measuring the visual quality of 2D images as it is far less complex than the current approach, which consists in performing global motion estimation to align the input images prior to applying the metrics, and offers better accuracy.

International Journal of Pattern Recognition and Artificial Intelligence, 2011

The denoising of a natural signal/image corrupted by Gaussian white noise is a classical problem ... more The denoising of a natural signal/image corrupted by Gaussian white noise is a classical problem in signal/image processing. However, it is still in its infancy to denoise high dimensional data. In this paper, we extended Sendur and Selesnick's bivariate wavelet thresholding from two-dimensional (2D) image denoising to three-dimensional (3D) data cube denoising. Our study shows that bivariate wavelet thresholding is still valid for 3D data cubes. Experimental results show that bivariate wavelet thresholding on 3D data cube is better than performing 2D bivariate wavelet thresholding on every spectral band separately, VisuShrink, and Chen and Zhu's 3-scale denoising.

2013 International Conference on High Performance Computing & Simulation (HPCS), 2013

Support vector machine (SVM) is one of the most popular classification techniques in pattern reco... more Support vector machine (SVM) is one of the most popular classification techniques in pattern recognition community. However, because of outliers in the training samples, SVM tends to perform poorly under such circumstances. In this paper, we borrow the idea from compressive sensing by introducing an extra term to the objective function of the standard SVM in order to achieve a sparse representation. Furthermore, instead of using the l 0 norm, we adopt the l 1 norm in our sparse SVM. In most cases, our method achieves higher classification rates than the standard SVM because of sparser support vectors and is more robust to outliers in the datasets. Experimental results show that our proposed SVM is efficient in pattern recognition applications.

Pattern Recognition and Image Analysis, 2013

ABSTRACT

Pattern Recognition, 2005

Image denoising by means of wavelet transforms has been an active research topic for many years. ... more Image denoising by means of wavelet transforms has been an active research topic for many years. For a given noisy image, which kind of wavelet and what threshold we use should have significant impact on the quality of the denoised image. In this paper, we use Simulated Annealing to find the customized wavelet filters and the customized threshold corresponding to the given noisy image at the same time. Also, we propose to consider a small neighbourhood around the customized wavelet coefficient to be thresholded for image denoising. Experimental results show that our approach is better than VisuShrink, our NeighShrink with fixed wavelet, and the wiener2 filter that is available in Matlab Image Processing Toolbox. In addition, our NeighShrink with fixed wavelet already outperforms VisuShrink for all the experiments.