Reza Safabakhsh - Academia.edu (original) (raw)

Uploads

Papers by Reza Safabakhsh

Research paper thumbnail of A novel medial axis based model for human eye description

Research paper thumbnail of Adaptive method for hiding data in images

Journal of Electronic Imaging, 2012

ABSTRACT Steganographic methods are desired to have the least impact on the statistical and perce... more ABSTRACT Steganographic methods are desired to have the least impact on the statistical and perceptual properties of the cover image to prevent steganalytical methods from the detection of data hiding in the image. The existing spatial domain methods of steganography generally have low data-hiding capacity and low security. We present an adaptive spatial domain steganographic method with increased security and capacity. The method partitions the image into flat and nonflat blocks and uses different embedding strategies based on the block type. These strategies are based on modifying certain pixels within each 3×3 block of the image. Experimental results show that the histogram and higher order statistics-based steganalytical methods cannot detect the stego images produced by the proposed method and that the capacity is increased significantly.

Research paper thumbnail of English to Persian transliteration using attention-based approach in deep learning

2017 Iranian Conference on Electrical Engineering (ICEE), 2017

In this paper, transliteration is carried out by using the attention-based approach in deep learn... more In this paper, transliteration is carried out by using the attention-based approach in deep learning. Unlike the previous works which randomly initialize the weights in the encoder, word vector representation of the source vocabulary has been used as an initial value for the weights. The representation is computed by counting the co-occurrences between different characters. Experimental results on an English to Persian transliteration corpus with more than 14000 word pairs show the superior performance of the proposed method (up to 4.21 BLEU points improvement) over the basic attention-based approach.

Research paper thumbnail of Graphical model based continuous estimation of distribution algorithm

Applied Soft Computing, 2017

Highlights  A new estimation of distribution algorithm based on learning mutual dependencies amo... more Highlights  A new estimation of distribution algorithm based on learning mutual dependencies among random variables using a probabilistic graphical model and sampling from it  Estimating the joint probability of selected solution using a GMM and parallel learning and sampling for each Gaussian mixture  Limiting the search space by using mutual dependencies for learning Markov network  Decomposing the search space of the network into the networks with a smaller number of nodes (subgraphs) by decomposing Markov network into the maximal cliques and the clique graph.  Parallel learning of the structure of a complete BN for each subgraph  Reducing the time complexity of learning a BN for modeling the joint probability distribution by using parallel learning in the subgraphs.  Considering more than three mutual dependencies among variables (complete BNs)  Proposing a score metric for learning a Bayesian network structure based on optimizing a constrained problem by determination of the information flow and uncertainty in the network  Sampling the offspring using ordered Cholesky decomposition

Research paper thumbnail of Traffic sign recognition using an extended bag-of-features model with spatial histogram

2015 Signal Processing and Intelligent Systems Conference (SPIS), 2015

Traffic sign recognition (TSR) is a major challenging task for intelligent transport systems. In ... more Traffic sign recognition (TSR) is a major challenging task for intelligent transport systems. In this paper, we present a multiclass traffic sign recognition system based on the Bag-of-Word (BOW) model. Despite huge success of BOW method, ignoring the spatial information is a weakness of this model and affects accuracy of classification. We have proposed a Spatial Histogram for traffic signs that preserves the required spatial information. In addition, we used an extended codebook construction method to extract key features from all of sign categories efficiently and achieved a recognition rate of %88.02 through 62 sign types with a short execution time.

Research paper thumbnail of Dynamics of the HPA axis and inflammatory cytokines: Insights from mathematical modeling

Computers in Biology and Medicine, 2015

In the work presented here, a novel mathematical model was developed to explore the bi-directiona... more In the work presented here, a novel mathematical model was developed to explore the bi-directional communication between the hypothalamic-pituitary-adrenal (HPA) axis and inflammatory cytokines in acute inflammation. The dynamic model consists of five delay differential equations 5D for two main pro-inflammatory cytokines (TNF-α and IL-6) and two hormones of the HPA axis (ACTH and cortisol) and LPS endotoxin. The model is an attempt to increase the understanding of the role of primary hormones and cytokines in this complex relationship by demonstrating the influence of different organs and hormones in the regulation of the inflammatory response. The model captures the main qualitative features of cytokine and hormone dynamics when a toxic challenge is introduced. Moreover, in this work a new simple delayed model of the HPA axis is introduced which supports the understanding of the ultradian rhythm of HPA hormones both in normal and infection conditions. Through simulations using the model, the role of key inflammatory cytokines and cortisol in transition from acute to persistent inflammation through stability analysis is investigated. Also, by employing a Markov chain Monte Carlo (MCMC) method, parameter uncertainty and the effects of parameter variations on each other are analyzed. This model confirms the important role of the HPA axis in acute and prolonged inflammation and can be a useful tool in further investigation of the role of stress on the immune response to infectious diseases.

Research paper thumbnail of Human Eye tracking via Composition of TASOM based ACM and Change Management Methods

Research paper thumbnail of Human eye boundary detection using a modified TASOM

Research paper thumbnail of Eye Segmentation via Maximum Contrast Difference Filtering

Research paper thumbnail of Crowd Density Estimation for Outdoor Environments

Proceedings of the 8th International Conference on Bio-inspired Information and Communications Technologies (formerly BIONETICS), 2015

ABSTRACT

Research paper thumbnail of Batch linear manifold topographic map with regional dimensionality estimation

2009 International Joint Conference on Neural Networks, 2009

ABSTRACT This paper introduces an unsupervised batch algorithm for learning the underlying region... more ABSTRACT This paper introduces an unsupervised batch algorithm for learning the underlying regional linear manifolds and estimating their dimensionalities using a data set in a topographic map. For this purpose, a unified free energy functional is designed and an expectation-maximization procedure is developed to minimize it. Regional dimensionality estimation controls the extent of the linear manifolds. This property makes the model appropriate for representing the datasets with varying regional intrinsic dimensions, compared to the resembling techniques without dimensionality learning capability. Experimental results show the good performance of the model on synthesized and realworld applications.

Research paper thumbnail of Evolution of neural network architecture and weights using mutation based genetic algorithm

2009 14th International CSI Computer Conference, 2009

ABSTRACT In this paper we present a new approach for evolving an optimized neural network archite... more ABSTRACT In this paper we present a new approach for evolving an optimized neural network architecture for a three layer feedforward neural network with a mutation based genetic algorithm. In this study we will optimize the weights and the network architecture simultaneously through a new presentation for the three layer feedforward neural network. The goal of the method is to find the optimal number of neurons and their appropriate weights. This optimization problem so far has been solved by looking at the general architecture of the network but we optimize the individual neurons of the hidden layer. This change results in a search space with much higher resolution and an increased speed of convergence. Evaluation of algorithm by 3 data sets reveals that this method shows a very good performance in comparison to current methods.

Research paper thumbnail of Paper: A TASOM-BASED ALGORITHM FOR ACTIVE CONTOUR MODELING

Research paper thumbnail of Automated visual inspection for detection, characterization, and measurement of flaws on machined metal surfaces (pattern recognition, wedge/ring, laser applications, avi)

Research paper thumbnail of Determining Differential Characteristics of Block Ciphers Using Hopfield Network and Boltzmann Machine

Research paper thumbnail of On Transformation Approach to Digital PID Controller Realization

Research paper thumbnail of The time adaptive self-organizing map with neighborhood functions for bilevel thresholding

Research paper thumbnail of Automatic adjustment of learning rates of the self-organizing feature map

Research paper thumbnail of NEFRL: A New Neuro-Fuzzy System for Episodic Reinforcement Learning Tasks

2007 Frontiers in the Convergence of Bioscience and Information Technologies, 2007

ABSTRACT In this paper, we propose a new neuro-fuzzy system for episodic reinforcement learning t... more ABSTRACT In this paper, we propose a new neuro-fuzzy system for episodic reinforcement learning tasks, NEFRL. While NEFRL has all benefits of a neuro-fuzzy architecture, it has the additional advantage that it can learn with a numerical evaluation of performance and there is no need for training input-output pairs. Also, we show that the learning algorithm of this system converges with probability one to a local maximum of the average numerical performance signal. Our experimental results for the pole-balancing task show the power of this system even without any prior domain knowledge.

Research paper thumbnail of A fast and accurate steganalysis using Ensemble classifiers

2013 8th Iranian Conference on Machine Vision and Image Processing (MVIP), 2013

ABSTRACT Nowadays the steganographic methods use the more sophisticated image models to increase ... more ABSTRACT Nowadays the steganographic methods use the more sophisticated image models to increase security; consequently, steganalysis algorithm should build the more accurate models of images to detect them. So, the number of extracted feature is increasing. Most modern steganalysis algorithms train a supervised classifier on the feature vectors. The most popular and accurate one is SVM, but the high training time of SVM inhibits the development of steganalysis. To solve this problem, in this paper we propose a fast and accurate steganalysis methods based on Ensemble classifier and Stacking. In this method, the relation between basic learners decisions and true decision is learned by another classifier. To do this, basic learners decisions are mapped to space of uncorrelated dimensions. The complexity of this method is much lower than that of SVM, while our method improves detection accuracy. Proposed method is a fast and accurate classifier that can be used as a part of any steganalysis algorithm. Performance of this method is demonstrated on two steganographic methods, namely nsF5 and Model Based Steganography. The performance of proposed method is compared to that of Ensemble classifier. Experimental results show that the classification error and training time are lowered by 46% and 88%, respectively.

Research paper thumbnail of A novel medial axis based model for human eye description

Research paper thumbnail of Adaptive method for hiding data in images

Journal of Electronic Imaging, 2012

ABSTRACT Steganographic methods are desired to have the least impact on the statistical and perce... more ABSTRACT Steganographic methods are desired to have the least impact on the statistical and perceptual properties of the cover image to prevent steganalytical methods from the detection of data hiding in the image. The existing spatial domain methods of steganography generally have low data-hiding capacity and low security. We present an adaptive spatial domain steganographic method with increased security and capacity. The method partitions the image into flat and nonflat blocks and uses different embedding strategies based on the block type. These strategies are based on modifying certain pixels within each 3×3 block of the image. Experimental results show that the histogram and higher order statistics-based steganalytical methods cannot detect the stego images produced by the proposed method and that the capacity is increased significantly.

Research paper thumbnail of English to Persian transliteration using attention-based approach in deep learning

2017 Iranian Conference on Electrical Engineering (ICEE), 2017

In this paper, transliteration is carried out by using the attention-based approach in deep learn... more In this paper, transliteration is carried out by using the attention-based approach in deep learning. Unlike the previous works which randomly initialize the weights in the encoder, word vector representation of the source vocabulary has been used as an initial value for the weights. The representation is computed by counting the co-occurrences between different characters. Experimental results on an English to Persian transliteration corpus with more than 14000 word pairs show the superior performance of the proposed method (up to 4.21 BLEU points improvement) over the basic attention-based approach.

Research paper thumbnail of Graphical model based continuous estimation of distribution algorithm

Applied Soft Computing, 2017

Highlights  A new estimation of distribution algorithm based on learning mutual dependencies amo... more Highlights  A new estimation of distribution algorithm based on learning mutual dependencies among random variables using a probabilistic graphical model and sampling from it  Estimating the joint probability of selected solution using a GMM and parallel learning and sampling for each Gaussian mixture  Limiting the search space by using mutual dependencies for learning Markov network  Decomposing the search space of the network into the networks with a smaller number of nodes (subgraphs) by decomposing Markov network into the maximal cliques and the clique graph.  Parallel learning of the structure of a complete BN for each subgraph  Reducing the time complexity of learning a BN for modeling the joint probability distribution by using parallel learning in the subgraphs.  Considering more than three mutual dependencies among variables (complete BNs)  Proposing a score metric for learning a Bayesian network structure based on optimizing a constrained problem by determination of the information flow and uncertainty in the network  Sampling the offspring using ordered Cholesky decomposition

Research paper thumbnail of Traffic sign recognition using an extended bag-of-features model with spatial histogram

2015 Signal Processing and Intelligent Systems Conference (SPIS), 2015

Traffic sign recognition (TSR) is a major challenging task for intelligent transport systems. In ... more Traffic sign recognition (TSR) is a major challenging task for intelligent transport systems. In this paper, we present a multiclass traffic sign recognition system based on the Bag-of-Word (BOW) model. Despite huge success of BOW method, ignoring the spatial information is a weakness of this model and affects accuracy of classification. We have proposed a Spatial Histogram for traffic signs that preserves the required spatial information. In addition, we used an extended codebook construction method to extract key features from all of sign categories efficiently and achieved a recognition rate of %88.02 through 62 sign types with a short execution time.

Research paper thumbnail of Dynamics of the HPA axis and inflammatory cytokines: Insights from mathematical modeling

Computers in Biology and Medicine, 2015

In the work presented here, a novel mathematical model was developed to explore the bi-directiona... more In the work presented here, a novel mathematical model was developed to explore the bi-directional communication between the hypothalamic-pituitary-adrenal (HPA) axis and inflammatory cytokines in acute inflammation. The dynamic model consists of five delay differential equations 5D for two main pro-inflammatory cytokines (TNF-α and IL-6) and two hormones of the HPA axis (ACTH and cortisol) and LPS endotoxin. The model is an attempt to increase the understanding of the role of primary hormones and cytokines in this complex relationship by demonstrating the influence of different organs and hormones in the regulation of the inflammatory response. The model captures the main qualitative features of cytokine and hormone dynamics when a toxic challenge is introduced. Moreover, in this work a new simple delayed model of the HPA axis is introduced which supports the understanding of the ultradian rhythm of HPA hormones both in normal and infection conditions. Through simulations using the model, the role of key inflammatory cytokines and cortisol in transition from acute to persistent inflammation through stability analysis is investigated. Also, by employing a Markov chain Monte Carlo (MCMC) method, parameter uncertainty and the effects of parameter variations on each other are analyzed. This model confirms the important role of the HPA axis in acute and prolonged inflammation and can be a useful tool in further investigation of the role of stress on the immune response to infectious diseases.

Research paper thumbnail of Human Eye tracking via Composition of TASOM based ACM and Change Management Methods

Research paper thumbnail of Human eye boundary detection using a modified TASOM

Research paper thumbnail of Eye Segmentation via Maximum Contrast Difference Filtering

Research paper thumbnail of Crowd Density Estimation for Outdoor Environments

Proceedings of the 8th International Conference on Bio-inspired Information and Communications Technologies (formerly BIONETICS), 2015

ABSTRACT

Research paper thumbnail of Batch linear manifold topographic map with regional dimensionality estimation

2009 International Joint Conference on Neural Networks, 2009

ABSTRACT This paper introduces an unsupervised batch algorithm for learning the underlying region... more ABSTRACT This paper introduces an unsupervised batch algorithm for learning the underlying regional linear manifolds and estimating their dimensionalities using a data set in a topographic map. For this purpose, a unified free energy functional is designed and an expectation-maximization procedure is developed to minimize it. Regional dimensionality estimation controls the extent of the linear manifolds. This property makes the model appropriate for representing the datasets with varying regional intrinsic dimensions, compared to the resembling techniques without dimensionality learning capability. Experimental results show the good performance of the model on synthesized and realworld applications.

Research paper thumbnail of Evolution of neural network architecture and weights using mutation based genetic algorithm

2009 14th International CSI Computer Conference, 2009

ABSTRACT In this paper we present a new approach for evolving an optimized neural network archite... more ABSTRACT In this paper we present a new approach for evolving an optimized neural network architecture for a three layer feedforward neural network with a mutation based genetic algorithm. In this study we will optimize the weights and the network architecture simultaneously through a new presentation for the three layer feedforward neural network. The goal of the method is to find the optimal number of neurons and their appropriate weights. This optimization problem so far has been solved by looking at the general architecture of the network but we optimize the individual neurons of the hidden layer. This change results in a search space with much higher resolution and an increased speed of convergence. Evaluation of algorithm by 3 data sets reveals that this method shows a very good performance in comparison to current methods.

Research paper thumbnail of Paper: A TASOM-BASED ALGORITHM FOR ACTIVE CONTOUR MODELING

Research paper thumbnail of Automated visual inspection for detection, characterization, and measurement of flaws on machined metal surfaces (pattern recognition, wedge/ring, laser applications, avi)

Research paper thumbnail of Determining Differential Characteristics of Block Ciphers Using Hopfield Network and Boltzmann Machine

Research paper thumbnail of On Transformation Approach to Digital PID Controller Realization

Research paper thumbnail of The time adaptive self-organizing map with neighborhood functions for bilevel thresholding

Research paper thumbnail of Automatic adjustment of learning rates of the self-organizing feature map

Research paper thumbnail of NEFRL: A New Neuro-Fuzzy System for Episodic Reinforcement Learning Tasks

2007 Frontiers in the Convergence of Bioscience and Information Technologies, 2007

ABSTRACT In this paper, we propose a new neuro-fuzzy system for episodic reinforcement learning t... more ABSTRACT In this paper, we propose a new neuro-fuzzy system for episodic reinforcement learning tasks, NEFRL. While NEFRL has all benefits of a neuro-fuzzy architecture, it has the additional advantage that it can learn with a numerical evaluation of performance and there is no need for training input-output pairs. Also, we show that the learning algorithm of this system converges with probability one to a local maximum of the average numerical performance signal. Our experimental results for the pole-balancing task show the power of this system even without any prior domain knowledge.

Research paper thumbnail of A fast and accurate steganalysis using Ensemble classifiers

2013 8th Iranian Conference on Machine Vision and Image Processing (MVIP), 2013

ABSTRACT Nowadays the steganographic methods use the more sophisticated image models to increase ... more ABSTRACT Nowadays the steganographic methods use the more sophisticated image models to increase security; consequently, steganalysis algorithm should build the more accurate models of images to detect them. So, the number of extracted feature is increasing. Most modern steganalysis algorithms train a supervised classifier on the feature vectors. The most popular and accurate one is SVM, but the high training time of SVM inhibits the development of steganalysis. To solve this problem, in this paper we propose a fast and accurate steganalysis methods based on Ensemble classifier and Stacking. In this method, the relation between basic learners decisions and true decision is learned by another classifier. To do this, basic learners decisions are mapped to space of uncorrelated dimensions. The complexity of this method is much lower than that of SVM, while our method improves detection accuracy. Proposed method is a fast and accurate classifier that can be used as a part of any steganalysis algorithm. Performance of this method is demonstrated on two steganographic methods, namely nsF5 and Model Based Steganography. The performance of proposed method is compared to that of Ensemble classifier. Experimental results show that the classification error and training time are lowered by 46% and 88%, respectively.