Pub Date : 2015-06-15DOI: 10.1109/AISP.2015.7123509
A. Dehghani, A. Pourmohammad
The problem of small target detection in infrared images is one of the most important areas of research in passive defense systems. This detection method is classified in the Electro optic systems group. Generally, the challenges of the field are divided into two parts: detection and tracking. 1) Due to long detection distance, the amplitude of target signal compared with heavy background clutter is weak. On the other hand, targets appear with few pixels, so that there is no obvious and usable structural and contextual information. 2) Another challenge in tracking small targets is partial obstruction or closeness of background's brightness level to brightness level of the desired target (fading). In this paper, first background is removed by subtracting row mean, then the target are tracking using morphological filtering, thresholding the identified targets and finally by Kalman filter.
{"title":"Small target detection and tracking based on the background elimination and Kalman filter","authors":"A. Dehghani, A. Pourmohammad","doi":"10.1109/AISP.2015.7123509","DOIUrl":"https://doi.org/10.1109/AISP.2015.7123509","url":null,"abstract":"The problem of small target detection in infrared images is one of the most important areas of research in passive defense systems. This detection method is classified in the Electro optic systems group. Generally, the challenges of the field are divided into two parts: detection and tracking. 1) Due to long detection distance, the amplitude of target signal compared with heavy background clutter is weak. On the other hand, targets appear with few pixels, so that there is no obvious and usable structural and contextual information. 2) Another challenge in tracking small targets is partial obstruction or closeness of background's brightness level to brightness level of the desired target (fading). In this paper, first background is removed by subtracting row mean, then the target are tracking using morphological filtering, thresholding the identified targets and finally by Kalman filter.","PeriodicalId":405857,"journal":{"name":"2015 The International Symposium on Artificial Intelligence and Signal Processing (AISP)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2015-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128332641","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 : 2015-03-03DOI: 10.1109/AISP.2015.7123491
B. Nasersharif, S. Abdali
A solution for separating speech from music signal as a single channel source separation is Non-negative Matrix Factorization (NMF). In this approach spectrogram of each source signal is factorized as multiplication of two matrices which are known as basis and weight matrices. To achieve proper estimation of signal spectrogram, weight and basis matrices are updated iteratively. To estimate distance between signal and its estimation a cost function is used usually. Different cost functions have been introduced based on Kullback-Leibler (KL) and Itakura-Saito (IS) divergences. IS divergence is scale-invariant and so it is suitable for the conditions in which the coefficients of signal have a large dynamic range, for example in music short-term spectra. Based on this IS property, in this paper, we propose to use IS divergence as cost function of NMF in the training stage for music and on the other hand we suggest to use KL divergence as NMF cost function in the training stage for speech. Moreover, in the decomposition stage, we propose to use a linear combination of these two divergences in addition to a regularization term which considers temporal continuity information as a prior knowledge. Experimental results on one hour of speech and music, shows a good trade-off between signal to inference ratio (SIR) of speech and music in comparison to conventional NMF methods.
{"title":"Speech/music separation using non-negative matrix factorization with combination of cost functions","authors":"B. Nasersharif, S. Abdali","doi":"10.1109/AISP.2015.7123491","DOIUrl":"https://doi.org/10.1109/AISP.2015.7123491","url":null,"abstract":"A solution for separating speech from music signal as a single channel source separation is Non-negative Matrix Factorization (NMF). In this approach spectrogram of each source signal is factorized as multiplication of two matrices which are known as basis and weight matrices. To achieve proper estimation of signal spectrogram, weight and basis matrices are updated iteratively. To estimate distance between signal and its estimation a cost function is used usually. Different cost functions have been introduced based on Kullback-Leibler (KL) and Itakura-Saito (IS) divergences. IS divergence is scale-invariant and so it is suitable for the conditions in which the coefficients of signal have a large dynamic range, for example in music short-term spectra. Based on this IS property, in this paper, we propose to use IS divergence as cost function of NMF in the training stage for music and on the other hand we suggest to use KL divergence as NMF cost function in the training stage for speech. Moreover, in the decomposition stage, we propose to use a linear combination of these two divergences in addition to a regularization term which considers temporal continuity information as a prior knowledge. Experimental results on one hour of speech and music, shows a good trade-off between signal to inference ratio (SIR) of speech and music in comparison to conventional NMF methods.","PeriodicalId":405857,"journal":{"name":"2015 The International Symposium on Artificial Intelligence and Signal Processing (AISP)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2015-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125789881","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 : 2015-03-03DOI: 10.1109/AISP.2015.7123490
Saeed Torabzadeh, Reza Safabaksh
A Comprehensive Database of Farsi printed texts is an essential resource for research in this area. Although there are some Arabic printed databases, but those databases do not have all the necessary features for Farsi or Arabic text recognition research. In this paper, we introduce a comprehensive Farsi printed text database called AUT-PFT. The purpose of this database is to provide a large-scale, real world, multi font and multi size corpus for training Farsi or Arabic text recognition systems. This database is made up of 10000 generated words. 127 unique glyphs are used in these words in a way that appearance distribution of glyphs is approximately uniform. These words are generated with 10 widely used Farsi fonts and 4 different font sizes. In order to have real world noise in this database, all generated images were printed and scanned. Ground truth data are also provided for this database and unlike other databases, detailed information about document text is provided at glyph level.
{"title":"AUT-PFT: A real world printed Farsi text image dataset","authors":"Saeed Torabzadeh, Reza Safabaksh","doi":"10.1109/AISP.2015.7123490","DOIUrl":"https://doi.org/10.1109/AISP.2015.7123490","url":null,"abstract":"A Comprehensive Database of Farsi printed texts is an essential resource for research in this area. Although there are some Arabic printed databases, but those databases do not have all the necessary features for Farsi or Arabic text recognition research. In this paper, we introduce a comprehensive Farsi printed text database called AUT-PFT. The purpose of this database is to provide a large-scale, real world, multi font and multi size corpus for training Farsi or Arabic text recognition systems. This database is made up of 10000 generated words. 127 unique glyphs are used in these words in a way that appearance distribution of glyphs is approximately uniform. These words are generated with 10 widely used Farsi fonts and 4 different font sizes. In order to have real world noise in this database, all generated images were printed and scanned. Ground truth data are also provided for this database and unlike other databases, detailed information about document text is provided at glyph level.","PeriodicalId":405857,"journal":{"name":"2015 The International Symposium on Artificial Intelligence and Signal Processing (AISP)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2015-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122863631","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 : 2015-03-03DOI: 10.1109/AISP.2015.7123507
Fatemeh Kowkabi, H. Ghassemian, A. Keshavarz
Linear combination of endmembers according to their abundance fractions at pixel level is as the result of low spatial resolution of hyperspectral sensors. Spectral unmixing problem is described by decomposing these medley pixels into a set of endmembers and their abundance fractions. Most of endmember extraction techniques are designed on the basis of spectral feature of images such as OSP. Also SSPP is implied which considers spatial content of image pixels besides spectral information. We propose a self-governing module prior the spectral based endmember extraction algorithms to achieve superior performance of RMSE and SAD-based errors by creating a new synthetic image using HYDRA tool and USGS library with various values of SNR in order to evaluate our method with OSP and SSPP+OSP. Experimental results in comparison with the mentioned methods show that the proposed method can unmix data more effectively.
{"title":"Spatial and spectral preprocessor for spectral mixture analysis of synthetic remotely sensed hyperspectral image","authors":"Fatemeh Kowkabi, H. Ghassemian, A. Keshavarz","doi":"10.1109/AISP.2015.7123507","DOIUrl":"https://doi.org/10.1109/AISP.2015.7123507","url":null,"abstract":"Linear combination of endmembers according to their abundance fractions at pixel level is as the result of low spatial resolution of hyperspectral sensors. Spectral unmixing problem is described by decomposing these medley pixels into a set of endmembers and their abundance fractions. Most of endmember extraction techniques are designed on the basis of spectral feature of images such as OSP. Also SSPP is implied which considers spatial content of image pixels besides spectral information. We propose a self-governing module prior the spectral based endmember extraction algorithms to achieve superior performance of RMSE and SAD-based errors by creating a new synthetic image using HYDRA tool and USGS library with various values of SNR in order to evaluate our method with OSP and SSPP+OSP. Experimental results in comparison with the mentioned methods show that the proposed method can unmix data more effectively.","PeriodicalId":405857,"journal":{"name":"2015 The International Symposium on Artificial Intelligence and Signal Processing (AISP)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2015-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122804857","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 : 2015-03-03DOI: 10.1109/AISP.2015.7123532
Hamed Nikbakht, H. Mirvaziri
Data clustering is a crucial technique in data mining that is used in many applications. In this paper, a new clustering algorithm based on gravitational search algorithm (GSA) and genetic operators is proposed. The local search solution is utilized throw the global search to avoid getting stock in local optima. The GSA is a new approach to solve optimization problem that inspired by Newtonian law of gravity. We compared the performances of the proposed method with some well-known clustering algorithms on five benchmark dataset from UCI Machine Learning Repository. The experimental results show that our approach outperforms other algorithms and has better solution in all datasets.
{"title":"A new algorithm for data clustering based on gravitational search algorithm and genetic operators","authors":"Hamed Nikbakht, H. Mirvaziri","doi":"10.1109/AISP.2015.7123532","DOIUrl":"https://doi.org/10.1109/AISP.2015.7123532","url":null,"abstract":"Data clustering is a crucial technique in data mining that is used in many applications. In this paper, a new clustering algorithm based on gravitational search algorithm (GSA) and genetic operators is proposed. The local search solution is utilized throw the global search to avoid getting stock in local optima. The GSA is a new approach to solve optimization problem that inspired by Newtonian law of gravity. We compared the performances of the proposed method with some well-known clustering algorithms on five benchmark dataset from UCI Machine Learning Repository. The experimental results show that our approach outperforms other algorithms and has better solution in all datasets.","PeriodicalId":405857,"journal":{"name":"2015 The International Symposium on Artificial Intelligence and Signal Processing (AISP)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2015-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114319496","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 : 2015-03-03DOI: 10.1109/AISP.2015.7123535
M. Hosseinzadeh, M. Eftekhari
In this paper, fuzzy C-means clustering and Rotation Forest (RF) are combined to construct a high performance classifier for imbalanced data classification. Data samples are clustered via fuzzy clustering and then fuzzy membership function matrix is added into data samples. Therefore, clusters memberships of samples are utilized as new features that are added into the original features. After that, RF is utilized for classification where the new set of features as well as the original ones are taken into account in the feature subspacing phase. The proposed algorithm utilizes SMOTE oversampling algorithm for balancing data samples. The obtained results confirm that our proposed method outperforms the other well-known bagging algorithms.
{"title":"Improving rotation forest performance for imbalanced data classification through fuzzy clustering","authors":"M. Hosseinzadeh, M. Eftekhari","doi":"10.1109/AISP.2015.7123535","DOIUrl":"https://doi.org/10.1109/AISP.2015.7123535","url":null,"abstract":"In this paper, fuzzy C-means clustering and Rotation Forest (RF) are combined to construct a high performance classifier for imbalanced data classification. Data samples are clustered via fuzzy clustering and then fuzzy membership function matrix is added into data samples. Therefore, clusters memberships of samples are utilized as new features that are added into the original features. After that, RF is utilized for classification where the new set of features as well as the original ones are taken into account in the feature subspacing phase. The proposed algorithm utilizes SMOTE oversampling algorithm for balancing data samples. The obtained results confirm that our proposed method outperforms the other well-known bagging algorithms.","PeriodicalId":405857,"journal":{"name":"2015 The International Symposium on Artificial Intelligence and Signal Processing (AISP)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2015-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127934087","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 : 2015-03-03DOI: 10.1109/AISP.2015.7123488
Razieh Hosseini, R. Azmi
Community detection in social network is a significant issue in the study of the structure of a network and understanding its characteristics. A community is a significant structure formed by nodes with more connections between them. In recent years, several algorithms have been presented for community detection in social networks among them label propagation algorithm is one of the fastest algorithms, but due to the randomness of the algorithm its performance is not suitable. In this paper, we propose an improved label propagation algorithm called memory-based label propagation algorithm (MLPA) for finding community structure in social networks. In the proposed algorithm, a simple memory element is designed for each node of graph and this element store the most frequent common adoption of labels iteratively. Our experiments on the standard social network datasets show a relative improvement in comparison with other community detection algorithms.
{"title":"Memory-based label propagation algorithm for community detection in social networks","authors":"Razieh Hosseini, R. Azmi","doi":"10.1109/AISP.2015.7123488","DOIUrl":"https://doi.org/10.1109/AISP.2015.7123488","url":null,"abstract":"Community detection in social network is a significant issue in the study of the structure of a network and understanding its characteristics. A community is a significant structure formed by nodes with more connections between them. In recent years, several algorithms have been presented for community detection in social networks among them label propagation algorithm is one of the fastest algorithms, but due to the randomness of the algorithm its performance is not suitable. In this paper, we propose an improved label propagation algorithm called memory-based label propagation algorithm (MLPA) for finding community structure in social networks. In the proposed algorithm, a simple memory element is designed for each node of graph and this element store the most frequent common adoption of labels iteratively. Our experiments on the standard social network datasets show a relative improvement in comparison with other community detection algorithms.","PeriodicalId":405857,"journal":{"name":"2015 The International Symposium on Artificial Intelligence and Signal Processing (AISP)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2015-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131886942","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 : 2015-03-03DOI: 10.1109/AISP.2015.7123537
M. K. Ebrahimpour, M. Eftekhari
In this paper, a Feature Selection (FS) method based on Hesitant Fuzzy Sets (HFS) is proposed. The ranking value of three filter methods (i.e. Fisher, Relief, Information Gain) for each feature are considered as Hesitant Fuzzy Elements (HFE) of that feature with respect to class relevancy, then hesitant correlation matrix of features is calculated. After that three similarity measures are considered to evaluate the second hesitant correlation matrix of features. The first correlation matrix represents the correlation of features with respect to their relevancy to the class. The second correlation matrix presents the correlation based on redundancy of features among themselves. One Hesitant Fuzzy Sets Clustering Algorithm (HFSCA) is run on these matrixes. Finally the intersection of clusters is considerd as a features subset which contains the highly relevance and lowly redundant features. The experimental results confirm the ability of our proposed method in both number of selected features and accuracy comparing to the other ones.
{"title":"Proposing a novel feature selection algorithm based on Hesitant Fuzzy Sets and correlation concepts","authors":"M. K. Ebrahimpour, M. Eftekhari","doi":"10.1109/AISP.2015.7123537","DOIUrl":"https://doi.org/10.1109/AISP.2015.7123537","url":null,"abstract":"In this paper, a Feature Selection (FS) method based on Hesitant Fuzzy Sets (HFS) is proposed. The ranking value of three filter methods (i.e. Fisher, Relief, Information Gain) for each feature are considered as Hesitant Fuzzy Elements (HFE) of that feature with respect to class relevancy, then hesitant correlation matrix of features is calculated. After that three similarity measures are considered to evaluate the second hesitant correlation matrix of features. The first correlation matrix represents the correlation of features with respect to their relevancy to the class. The second correlation matrix presents the correlation based on redundancy of features among themselves. One Hesitant Fuzzy Sets Clustering Algorithm (HFSCA) is run on these matrixes. Finally the intersection of clusters is considerd as a features subset which contains the highly relevance and lowly redundant features. The experimental results confirm the ability of our proposed method in both number of selected features and accuracy comparing to the other ones.","PeriodicalId":405857,"journal":{"name":"2015 The International Symposium on Artificial Intelligence and Signal Processing (AISP)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2015-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130789042","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 : 2015-03-03DOI: 10.1109/AISP.2015.7123498
Reza Refaei Afshar, M. Asadpour
This paper propose a new approach to predict spreading behavior of conventions. Conventions in our case are verbal i.e. phrases used by many people for a new purpose regarding a social issue. We study usage of some conventions in Twitter popularized among Persian speaking users. We show that the number of tweets that contain a convention phrase in a period has a bell shaped curve. We use the latent space model to calculate the distance matrix for a convention in order to understand its spreading behavior. We first calculate the distance matrices of the conventions and utilize them to estimate the distance matrix for new conventions.
{"title":"Latent space model for analysis of conventions","authors":"Reza Refaei Afshar, M. Asadpour","doi":"10.1109/AISP.2015.7123498","DOIUrl":"https://doi.org/10.1109/AISP.2015.7123498","url":null,"abstract":"This paper propose a new approach to predict spreading behavior of conventions. Conventions in our case are verbal i.e. phrases used by many people for a new purpose regarding a social issue. We study usage of some conventions in Twitter popularized among Persian speaking users. We show that the number of tweets that contain a convention phrase in a period has a bell shaped curve. We use the latent space model to calculate the distance matrix for a convention in order to understand its spreading behavior. We first calculate the distance matrices of the conventions and utilize them to estimate the distance matrix for new conventions.","PeriodicalId":405857,"journal":{"name":"2015 The International Symposium on Artificial Intelligence and Signal Processing (AISP)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2015-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114250584","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 : 2015-03-03DOI: 10.1109/AISP.2015.7123534
K. Aghajani, Mohsen Shirpour, M. T. Manzuri
Image registration is an important task in medical image processing. Assuming spatially stationary intensity relation among images, conventional area based algorithms such as CC (Correlation Coefficients) and MI (Mutual Information), show weaker results alongside spatially varying intensity distortion. In this research, a structural representation of images is introduced. It allows us to use simpler similarity metrics in multimodal images which are also robust against the mentioned distortion field. The efficiency of this presentation in non-rigid image registration in the presence of spatial varying distortion field is examined. Experimental results on synthetic and real-world data sets demonstrate the effectiveness of the proposed method for image registration tasks.
{"title":"Structural image representation for image registration","authors":"K. Aghajani, Mohsen Shirpour, M. T. Manzuri","doi":"10.1109/AISP.2015.7123534","DOIUrl":"https://doi.org/10.1109/AISP.2015.7123534","url":null,"abstract":"Image registration is an important task in medical image processing. Assuming spatially stationary intensity relation among images, conventional area based algorithms such as CC (Correlation Coefficients) and MI (Mutual Information), show weaker results alongside spatially varying intensity distortion. In this research, a structural representation of images is introduced. It allows us to use simpler similarity metrics in multimodal images which are also robust against the mentioned distortion field. The efficiency of this presentation in non-rigid image registration in the presence of spatial varying distortion field is examined. Experimental results on synthetic and real-world data sets demonstrate the effectiveness of the proposed method for image registration tasks.","PeriodicalId":405857,"journal":{"name":"2015 The International Symposium on Artificial Intelligence and Signal Processing (AISP)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2015-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133433052","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：