Pub Date : 2002-11-07DOI: 10.1109/NNSP.2002.1030096
T. Kolenda, L. K. Hansen, J. Larsen, O. Winther
Independent component analysis of combined text and image data from Web pages has potential for search and retrieval applications by providing more meaningful and context dependent content. It is demonstrated that ICA of combined text and image features has a synergistic effect, i.e., the retrieval classification rates increase if based on multimedia components relative to single media analysis. For this purpose a simple probabilistic supervised classifier which works from unsupervised ICA features is invoked. In addition, we demonstrate the suggested framework for automatic annotation of descriptive key words to images.
{"title":"Independent component analysis for understanding multimedia content","authors":"T. Kolenda, L. K. Hansen, J. Larsen, O. Winther","doi":"10.1109/NNSP.2002.1030096","DOIUrl":"https://doi.org/10.1109/NNSP.2002.1030096","url":null,"abstract":"Independent component analysis of combined text and image data from Web pages has potential for search and retrieval applications by providing more meaningful and context dependent content. It is demonstrated that ICA of combined text and image features has a synergistic effect, i.e., the retrieval classification rates increase if based on multimedia components relative to single media analysis. For this purpose a simple probabilistic supervised classifier which works from unsupervised ICA features is invoked. In addition, we demonstrate the suggested framework for automatic annotation of descriptive key words to images.","PeriodicalId":117945,"journal":{"name":"Proceedings of the 12th IEEE Workshop on Neural Networks for Signal Processing","volume":"29 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126139770","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 : 2002-11-07DOI: 10.1109/NNSP.2002.1030079
Wei-Chen Chang, A. Su
Struck string instruments, such as pianos, usually have groups of strings with each group terminated at a common bridge. Because of the strong coupling phenomenon, the produced tones exhibit highly complex amplitude modulation patterns. Therefore, it is difficult to determine synthesis model parameters such that the synthesized tones can match recorded tones. A multi-channel recurrent network is proposed based on three previous works: the coupled-string model, the commuted piano synthesis method and the IIR synthesis method. This work attempts to extract automatically the synthesis parameters by using a neural-network training algorithm without the knowledge of the physical properties of the instruments. Computer simulations show encouraging results.
{"title":"A multi-channel recurrent network for synthesizing struck coupled-string musical instruments","authors":"Wei-Chen Chang, A. Su","doi":"10.1109/NNSP.2002.1030079","DOIUrl":"https://doi.org/10.1109/NNSP.2002.1030079","url":null,"abstract":"Struck string instruments, such as pianos, usually have groups of strings with each group terminated at a common bridge. Because of the strong coupling phenomenon, the produced tones exhibit highly complex amplitude modulation patterns. Therefore, it is difficult to determine synthesis model parameters such that the synthesized tones can match recorded tones. A multi-channel recurrent network is proposed based on three previous works: the coupled-string model, the commuted piano synthesis method and the IIR synthesis method. This work attempts to extract automatically the synthesis parameters by using a neural-network training algorithm without the knowledge of the physical properties of the instruments. Computer simulations show encouraging results.","PeriodicalId":117945,"journal":{"name":"Proceedings of the 12th IEEE Workshop on Neural Networks for Signal Processing","volume":"46 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129238913","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 : 2002-11-07DOI: 10.1109/NNSP.2002.1030092
Jiang-Yao Chen, Shun-Hsyung Chang, S. Leu
An adaptive back propagation (BP) neural network based PN code acquisition system is presented. Conventional neural network based acquisition systems are usually trained on PN code, but this system is based on training a back propagation neural network at all possible phases of the output of a correlation detector which is modified by a recursive accumulator. The recursive accumulator can converge the input of the neural network into a limited sample space, and the BP neural network acquires the phase of the received PN code from the converged data. The advantages of this system are that the gain of the system is controllable and the training data sample space is limited. The BP neural network is used to distinguish the transmitted signal and noise. Computer simulations show that the proposed system can acquire the phase of the received PN code correctly at very low signal-to-noise ratio (SNR) in an AWGN channel.
{"title":"Adaptive BP neural network (ABPNN) based PN code acquisition system via recursive accumulator","authors":"Jiang-Yao Chen, Shun-Hsyung Chang, S. Leu","doi":"10.1109/NNSP.2002.1030092","DOIUrl":"https://doi.org/10.1109/NNSP.2002.1030092","url":null,"abstract":"An adaptive back propagation (BP) neural network based PN code acquisition system is presented. Conventional neural network based acquisition systems are usually trained on PN code, but this system is based on training a back propagation neural network at all possible phases of the output of a correlation detector which is modified by a recursive accumulator. The recursive accumulator can converge the input of the neural network into a limited sample space, and the BP neural network acquires the phase of the received PN code from the converged data. The advantages of this system are that the gain of the system is controllable and the training data sample space is limited. The BP neural network is used to distinguish the transmitted signal and noise. Computer simulations show that the proposed system can acquire the phase of the received PN code correctly at very low signal-to-noise ratio (SNR) in an AWGN channel.","PeriodicalId":117945,"journal":{"name":"Proceedings of the 12th IEEE Workshop on Neural Networks for Signal Processing","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130955696","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 : 2002-11-07DOI: 10.1109/NNSP.2002.1030053
E. McDermott, S. Katagiri
This article shows that the minimum classification error (MCE) criterion function commonly used for discriminative design of pattern recognition systems is equivalent to a Parzen window based estimate of the theoretical Bayes classification risk. In this analysis, each training token is mapped to the center of a Parzen kernel in the domain of a suitably defined "output level" random variable. The kernels are summed to produce a density estimate; this estimate in turn can easily be integrated over the domain of incorrect classifications, yielding the risk estimate. The expression of risk for each kernel can be seen to correspond directly to the usual MCE loss function. The resulting risk estimate can be minimized by suitable adaptation of the recognition system parameters that determine the mapping from training token to kernel center. This analysis provides a novel link between the MCE empirical cost measured on a finite training set and the theoretical Bayes classification risk.
{"title":"Minimum classification error via a Parzen window based estimate of the theoretical Bayes classification risk","authors":"E. McDermott, S. Katagiri","doi":"10.1109/NNSP.2002.1030053","DOIUrl":"https://doi.org/10.1109/NNSP.2002.1030053","url":null,"abstract":"This article shows that the minimum classification error (MCE) criterion function commonly used for discriminative design of pattern recognition systems is equivalent to a Parzen window based estimate of the theoretical Bayes classification risk. In this analysis, each training token is mapped to the center of a Parzen kernel in the domain of a suitably defined \"output level\" random variable. The kernels are summed to produce a density estimate; this estimate in turn can easily be integrated over the domain of incorrect classifications, yielding the risk estimate. The expression of risk for each kernel can be seen to correspond directly to the usual MCE loss function. The resulting risk estimate can be minimized by suitable adaptation of the recognition system parameters that determine the mapping from training token to kernel center. This analysis provides a novel link between the MCE empirical cost measured on a finite training set and the theoretical Bayes classification risk.","PeriodicalId":117945,"journal":{"name":"Proceedings of the 12th IEEE Workshop on Neural Networks for Signal Processing","volume":"411 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126690117","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 : 2002-11-07DOI: 10.1109/NNSP.2002.1030064
Michel Haritopoulos, Hujun Yin, N. Allinson
We treat self-organizing maps (SOMs) as means for denoising of images corrupted by multiplicative noise. To achieve this goal, we propose a scheme for blind source separation based on a nonlinear topology preserving mapping as it is performed by SOMs. Despite the assumption that only two noisy frames of the same image scene are available, we show that by a suitable post-processing step based on the estimates provided by the SOM, one can obtain enhanced versions of the originally noisy scenes. Our work is illustrated by application results of the proposed method to test and real images.
{"title":"Self-organizing map applied to image denoising","authors":"Michel Haritopoulos, Hujun Yin, N. Allinson","doi":"10.1109/NNSP.2002.1030064","DOIUrl":"https://doi.org/10.1109/NNSP.2002.1030064","url":null,"abstract":"We treat self-organizing maps (SOMs) as means for denoising of images corrupted by multiplicative noise. To achieve this goal, we propose a scheme for blind source separation based on a nonlinear topology preserving mapping as it is performed by SOMs. Despite the assumption that only two noisy frames of the same image scene are available, we show that by a suitable post-processing step based on the estimates provided by the SOM, one can obtain enhanced versions of the originally noisy scenes. Our work is illustrated by application results of the proposed method to test and real images.","PeriodicalId":117945,"journal":{"name":"Proceedings of the 12th IEEE Workshop on Neural Networks for Signal Processing","volume":"30 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126774594","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 : 2002-11-07DOI: 10.1109/NNSP.2002.1030048
A. G. Ramakrishnan, S. Raja, H. Ram
The effectiveness of Gabor filters for texture segmentation is well known. In this paper, we propose a texture identification scheme, based on a neural network (NN) using Gabor features. The features are derived from both the Gabor cosine and sine filters. Through experiments, we demonstrate the effectiveness of a NN based classifier using Gabor features for identifying textures in a controlled environment. The neural network used for texture identification is based on the multilayer perceptron (MLP) architecture. The classification results obtained show an improvement over those obtained by K-means clustering and maximum likelihood approaches.
{"title":"Neural network-based segmentation of textures using Gabor features","authors":"A. G. Ramakrishnan, S. Raja, H. Ram","doi":"10.1109/NNSP.2002.1030048","DOIUrl":"https://doi.org/10.1109/NNSP.2002.1030048","url":null,"abstract":"The effectiveness of Gabor filters for texture segmentation is well known. In this paper, we propose a texture identification scheme, based on a neural network (NN) using Gabor features. The features are derived from both the Gabor cosine and sine filters. Through experiments, we demonstrate the effectiveness of a NN based classifier using Gabor features for identifying textures in a controlled environment. The neural network used for texture identification is based on the multilayer perceptron (MLP) architecture. The classification results obtained show an improvement over those obtained by K-means clustering and maximum likelihood approaches.","PeriodicalId":117945,"journal":{"name":"Proceedings of the 12th IEEE Workshop on Neural Networks for Signal Processing","volume":"42 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127458626","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 : 2002-11-07DOI: 10.1109/NNSP.2002.1030060
H. Snoussi, A. Mohammad-Djafari
We consider the problem of the blind separation of noisy instantaneously mixed images. The images are modelized by hidden Markov fields with unknown parameters. Given the observed images, we give a Bayesian formulation and we propose to solve the resulting data augmentation problem by implementing a Monte Carlo Markov chain (MCMC) procedure. We separate the unknown variables into two categories: (1) the parameters of interest which are the mixing matrix, the noise covariance and the parameters of the sources distributions; and (2) the hidden variables which are the unobserved sources and the unobserved pixels classification labels. The proposed algorithm provides in the stationary regime samples drawn from the posterior distributions of all the variables involved in the problem leading to a flexibility in the cost function choice. We discuss and characterize some problems of non-identifiability and degeneracies of the parameters likelihood and the behavior of the MCMC algorithm in this case. Finally, we show the results for both synthetic and real data to illustrate the feasibility of the proposed solution.
{"title":"MCMC joint separation and segmentation of hidden Markov fields","authors":"H. Snoussi, A. Mohammad-Djafari","doi":"10.1109/NNSP.2002.1030060","DOIUrl":"https://doi.org/10.1109/NNSP.2002.1030060","url":null,"abstract":"We consider the problem of the blind separation of noisy instantaneously mixed images. The images are modelized by hidden Markov fields with unknown parameters. Given the observed images, we give a Bayesian formulation and we propose to solve the resulting data augmentation problem by implementing a Monte Carlo Markov chain (MCMC) procedure. We separate the unknown variables into two categories: (1) the parameters of interest which are the mixing matrix, the noise covariance and the parameters of the sources distributions; and (2) the hidden variables which are the unobserved sources and the unobserved pixels classification labels. The proposed algorithm provides in the stationary regime samples drawn from the posterior distributions of all the variables involved in the problem leading to a flexibility in the cost function choice. We discuss and characterize some problems of non-identifiability and degeneracies of the parameters likelihood and the behavior of the MCMC algorithm in this case. Finally, we show the results for both synthetic and real data to illustrate the feasibility of the proposed solution.","PeriodicalId":117945,"journal":{"name":"Proceedings of the 12th IEEE Workshop on Neural Networks for Signal Processing","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128343937","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 : 2002-11-07DOI: 10.1109/NNSP.2002.1030065
A. Cichocki, P. Georgiev
In many applications of independent component analysis (ICA) and blind source separation (BSS) the mixing or separating matrices have some special structure or some constraints are imposed for the matrices like symmetry, orthogonality, nonnegativity, sparseness and unit (or specified invariant norm) of the matrix. We present several algorithms and overview some known transformations which allows us to preserve such constraints. Especially, we propose algorithms for a blind identification problem with non-negativity constraints.
{"title":"Blind identification problems with constraints","authors":"A. Cichocki, P. Georgiev","doi":"10.1109/NNSP.2002.1030065","DOIUrl":"https://doi.org/10.1109/NNSP.2002.1030065","url":null,"abstract":"In many applications of independent component analysis (ICA) and blind source separation (BSS) the mixing or separating matrices have some special structure or some constraints are imposed for the matrices like symmetry, orthogonality, nonnegativity, sparseness and unit (or specified invariant norm) of the matrix. We present several algorithms and overview some known transformations which allows us to preserve such constraints. Especially, we propose algorithms for a blind identification problem with non-negativity constraints.","PeriodicalId":117945,"journal":{"name":"Proceedings of the 12th IEEE Workshop on Neural Networks for Signal Processing","volume":"43 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121354475","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 : 2002-11-07DOI: 10.1109/NNSP.2002.1030038
Christos Boukis, D. Mandic, A. Constantinides
An architecture for adaptive filtering based upon the previously introduced hierarchical least mean square algorithm is proposed. This pyramidal architecture incorporates sparse connections between the architectural layers with a certain variable degree of overlapping between the neighboring subfilters of the same level. A learning algorithm for this class of structures is derived, based on the back-propagation algorithm for temporal feedforward networks with linear neurons. Further, a class of normalized algorithms for this class is derived. The analysis and simulations show the proposed algorithms outperform the existing ones.
{"title":"A hierarchical feedforward adaptive filter for system identification","authors":"Christos Boukis, D. Mandic, A. Constantinides","doi":"10.1109/NNSP.2002.1030038","DOIUrl":"https://doi.org/10.1109/NNSP.2002.1030038","url":null,"abstract":"An architecture for adaptive filtering based upon the previously introduced hierarchical least mean square algorithm is proposed. This pyramidal architecture incorporates sparse connections between the architectural layers with a certain variable degree of overlapping between the neighboring subfilters of the same level. A learning algorithm for this class of structures is derived, based on the back-propagation algorithm for temporal feedforward networks with linear neurons. Further, a class of normalized algorithms for this class is derived. The analysis and simulations show the proposed algorithms outperform the existing ones.","PeriodicalId":117945,"journal":{"name":"Proceedings of the 12th IEEE Workshop on Neural Networks for Signal Processing","volume":"89 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128750105","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 : 2002-11-07DOI: 10.1109/NNSP.2002.1030061
T. Ristaniemi, K. Raju, J. Karhunen, E. Oja
In this paper independent component analysis (ICA) is considered for blind interference cancellation in a direct sequence spread spectrum communication system utilizing antenna arrays. An ICA-assisted interference canceler was proposed by Ristaniemi, Raju and Karhunen (see Proc. IEEE Int. Conf. on Communications, New York, USA, April 2002). This receiver structure is an extension to the framework proposed by Belouchrani and Amin (see Signal Processing, vol.80, p.723-29, 2000), in which blind source separation (BSS) techniques were utilized to the jammer mitigation problem. A common feature for both is that they apply an advanced pre-processing tool to offer an unjammed signal for conventional detection. However, it is not always desirable to apply the pre-processing tool, since it might even cause additional interference if the jammer is weak or absent. What would make the receivers more practical is to switch the additional canceler active only whenever it is expected to improve conventional detection. We compare two possible switching strategies at both ends of the receiver chain, pre- and post-switching schemes, and evaluate their impacts to the overall performance improvement of the array receiver.
{"title":"Jammer cancellation in DS-CDMA arrays: pre and post switching of ICA and RAKE","authors":"T. Ristaniemi, K. Raju, J. Karhunen, E. Oja","doi":"10.1109/NNSP.2002.1030061","DOIUrl":"https://doi.org/10.1109/NNSP.2002.1030061","url":null,"abstract":"In this paper independent component analysis (ICA) is considered for blind interference cancellation in a direct sequence spread spectrum communication system utilizing antenna arrays. An ICA-assisted interference canceler was proposed by Ristaniemi, Raju and Karhunen (see Proc. IEEE Int. Conf. on Communications, New York, USA, April 2002). This receiver structure is an extension to the framework proposed by Belouchrani and Amin (see Signal Processing, vol.80, p.723-29, 2000), in which blind source separation (BSS) techniques were utilized to the jammer mitigation problem. A common feature for both is that they apply an advanced pre-processing tool to offer an unjammed signal for conventional detection. However, it is not always desirable to apply the pre-processing tool, since it might even cause additional interference if the jammer is weak or absent. What would make the receivers more practical is to switch the additional canceler active only whenever it is expected to improve conventional detection. We compare two possible switching strategies at both ends of the receiver chain, pre- and post-switching schemes, and evaluate their impacts to the overall performance improvement of the array receiver.","PeriodicalId":117945,"journal":{"name":"Proceedings of the 12th IEEE Workshop on Neural Networks for Signal Processing","volume":"23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-11-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123729980","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：