Particle Swarm Optimization on Parallel Computers for Improving the Performance of a Gait Recognition System

Journal Polytechnic Journal

publisher Polytechnic Journal

Volume Volume 9

Read Publication

Human gait recognition based on feature extraction of support vector machine and pattern network algorithm

Journal IOP Conference Series: Materials Science and Engineering

DOI 10.1088/1757-899X/518/5/052010

Issue Issue 5

Volume Volume 51

Human recognition based on biometric information is important due to its reliability in identity verification. Gait recognition has ability to recognize individuals from a distance. "This study includes human gait recognition based firstly on support vector machine (SVM) and secondly on PatternNet neural network". "Three feature extraction and dimension reduction algorithms were used to increase the recognition performance of these algorithms". These algorithms are: "Liner Discriminant Analysis (LDA), Discrete Fourier Transform (DFT)and Discrete Cosine Transform (DCT)".The performances were compared using mean square error (MSE), PSNR and recognition rate to identify the best model and algorithm. The best results were obtained from the patternNet model especially when it was trained with TrainLM were correct classification rates (CCR) (98%), MSE (0.001) and PSNR (42) where "obtained when adopting LDA algorithm in comparison with DFT and DCT".

Read Publication

Support Vector Machine vs Pattern Network for Gait Recognition

Human recognition based on biometric information is important due to its reliability in identity verification. Gait recognition has ability to recognize individuals from a distance. This study includes human gait recognition based firstly on support vector machine (SVM) and secondly on PatternNet neural network. Experiments were conducted with comparisons based on the two approaches. Experimental results showed that the PattenNet neural network is more effective than the SVM in gait recognition.

Read Publication

Particle Swarm Optimization Based Discrete Cosine Transform for Person Identification by Gait Recognition

Journal The 7th International Conference on Information Technology (ICIT 2015), AlZaytoonah University of Jordan, Amman, Jordan

DOI 10.15849/icit.2015.0023

Gait recognition addresses the problem of human identification at a distance by identifying people based on the way they walk. Therefore, gait recognition has gained growing interest from researchers in recent years. This work presents gait recognition system based on particle swarm optimization (PSO) to recognize a person performing the movement for person identification. The system is based on Discrete Cosine Transform (DCT) for reducing dimensionality and feature extraction. Many experiments were conducted using different: swarm size, block dimension and number of iterations. The results showed that increasing the swarm size to 40 particles and also increasing block size of DCT sub image to (70×70) pixels will increase the overall performance of gait recognition system. The recognition rate reached 96%, MSE reached 0.0088 and PSNR reached 35%.

Read Publication

Performance Analysis of Different Feature Extraction Algorithms Used with Particle Swarm Optimization for Gait Recognition System

Journal International Journal of Recent Technology and Engineering (IJRTE)

Issue Issue-2

Volume Volume 4

Recently, person identification systems based on gait recognition had been gained growing large interest from researchers in the fields of artificial intelligence and image processing Thus, a gait recognition system based on particle swarm optimization (PSO) has been suggested in this work to recognize any person at a distance who performing the movement. Three feature extraction and dimension reduction algorithms were used to increase the recognition performance of PSO algorithm. These algorithms are: Liner Discriminant Analysis (LDA); Discrete Fourier Transform (DFT); and Discrete Cosine Transform (DCT). Many experiments were conducted for PSO with the three algorithms using different: swarm size, block dimension and number of iterations. Best results obtained when selecting swarm size equal 40, feature block size 70×70 and 100 number of iterations. At the same time best results of: recognition rate (97%), MSE (0.0027) and PSNR (38) where obtained when adopting LDA algorithm in comparison with DFT and DCT. And also the results obtained from DFT are better than the results obtained from using DCT. The time required for executing the LDA is lowest than the time required for executing DFT and DCT. DCT require more time than the other used feature extraction algorithms.

Read Publication

Pattern Recognition Neural Network for Improving the Performance of Iris Recognition System

Journal International Journal of Scientific & Engineering Research,

Issue Issue 6

Volume Volume 4

This research employs pattern recognition neural networks for Iris recognition systems. The neural networks have a sevenlayers architecture consisting of one input layer, five hidden layers, and one output layer. Ten different ANN optimization training algorithms were used separately to train this model to get best results for the iris recognition system. Many experiments were conducted to compare the results of this model with the results of other ANN models to identify the model which improves the performance of the iris recognition system. The performances were compared using mean square error (MSE), PSNR and recognition rate to identify the best model and algorithm. The best results were obtained from the patternNet model especially when it was trained with TrainLM. The results of this model were compared also with the results of other researches to show its efficiency

Read Publication

Using Parallel Genetic Algorithms to Compress Fractal Images

Journal Journal of Prospective Researches

publisher Polytechnic Journal

Volume Volume 4

Efficient technologies have been recently used in fractal image coding (FIC) to reduce the complexity of searching for matching between range block and domain block. The research aims at using the Parallel Genetic Algorithm (PGA) by the technology of the (Manager/Worker) in parallel computers to obtain matched domain blocks that prevent unsuitable convergence by coding the site of the searching domain block with a Gray code and a fitness function that minimizes the space between the matching of the current range block with the domain block under discussion in order to choose a protection strategy and coding of high accuracy for any image. Results showed that PGA is successful in fractal image coding and is flexible and efficient in reaching the optimum solution in higher speed and efficiency through using the Gray code. The searching method used for the parallel algorithm for compression and decompression, the method of choosing GA's coefficients, the selection, the crossover and mutation had a significant role in improving the image compression ratio and quality. Compression ratio has reached 87% while the image quality was improved after decompression that reached roughly 33% compared to traditional method in fractal image coding (FIC). Keywords-Parallel Genetic Algorithm (PGA), Fractal Image Coding (FIC), Local Iterated Function System (LIFS), Rang Block, Domain Block

Read Publication

نظام هجين: توازي خوارزمية جينية-عصبية في كبس الصور الكسوري باستخدام حاسبات متعددة

Journal AL-Rafidain Journal of Computer Sciences and Mathematics

publisher Polytechnic Journal

Volume Volume 10

Recently, effective technologies in Fractal Image Coding (FIC) were used to reduce the complexity of search for the matching between the Range blocks and the Domain blocks which reduces the time needed for calculation. The aim of this research is to propose a Hybird Parallel Neural-Genetic Algorithm (HPNGA) using the technique of (Manager/Worker) in multiple computers in order to obtain the fastest and best compression through extracting the features of the gray and colored images to attenuate the problem of dimensions in them. The NN enabled to train separate images from the test images to reduce the calculation time. The NN able to adapt itself with the training data to reduce the complexity and having more data and is merged with the parallel GA to reach optimum values of weights with their biases. The optimum weights obtained will classify the correct search domains with the least deviation, which, in turn, helps decompress the images using the fractal method with the minimum time and with high resolution through multiple computers. The results showed that the proposed hybrid system is faster than the standard algorithm, the NN and GA in decompressing the FIC and they are flexible and effective to reach the optimum solution with high speed and resolution. The search method used for compression and de-compression has a vital role in improving the ratio and the quality of image compression which reached 15s. The ratio of compression reached to 90.68% and the image improvement after decompression reached to 34.71 db when compared to other methods of (FIC), which didn't exceed 90.41% and image quality of 32.41

Read Publication

Artificial neural networks for iris recognition system: comparisons between different models, architectures and algorithms

Journal International journal of information and communication technology research

Volume Volume 10

Read Publication

Genetic algorithm based on parallel computing to improve the performance of fractal image compression system

Journal Eur. J. Sci. Res

Volume Volume 92

Read Publication

Modeling Soil Temperature at different depths and times as a function of some climatic data Using Artificial Neural Network

Journal journal of kerbala university

publisher Kerbala University

In this study, implementation of artificial neural network model has been used to estimate soil temperatures at various depths and different measuring times, as a function of mean air temperature, number of sunshine hours, radiation, for any day of the year. ANN (artificial neural network) of back propagation and fitness algorithms models. The data of soil temperature is taken from research department of soil and water/Nineveh province for the period from 1980 to 1983 and it include daily measurements of soil at depths of 5, 10, 20, 30, 50 and 100 cm and for three periods at 9, 12 and 15 clock for cultivated and bare soil. The data of two years was used to learn the network and the data of one year was used to test the network and compare its output with the measured data, three performance functions, namely root mean square errors (RMSE) and determination coefficient (R2), were used to evaluate the neural model, to find the adequacy between estimated data and the outputs of neural network for one year, the values of R2 ranging between 0.95-0.99 and the values of RMSE decreased significantly for all cases of estimation. The results shows the possibility of using neural networks in the composition of the model that can be used in the estimation of deep soil temperatures through the use of surface soil temperature for three times of measurement, the successful use of neural networks in the composition of the model that can be used to estimate the deep soil temperatures through the use of soil-surface temperatures, which are measured at different time periods. Successful construction of General ANN model that predict soil temperature at any …

Read Publication

استخدام الخوارزميات الجينية المتوازية للكشف عن البيانات المتطرفة حسب مقياس BIC

Journal مجلة تكريت للعلوم الصرفة

publisher جامعة تكريت

Read Publication

Backpropagation Neural Network Algorithm for Forecasting Soil Temperatures Considering Many Aspects: A Comparison of Different Approaches

Journal ICIT 2011 The 5th International Conference on Information Technology

Artificial Neural Networks (ANNs) are interconnected collections of processing units which have been used in different applications. The objective of this paper is to develop an ANN model to estimate soil temperature for any day by using various previous day meteorological variables. For this purpose, average temperature of air, sunshine, radiation and soil temperature for meteorological data between the years of 1980 and 1984 at Nineveh/Iraq Meteorological Station were used. We measured the soil temperatures at different depths of 5, 10, 20, 50 and 100cm within the time 9, 12 and 15 respectively. Three ANNs models were constructed. The Backpropagation neural network algorithm (BP), Cascade-Forward and Time Series (or Nonlinear Autoregressive) algorithms were used for the training the constructed ANNs models. These constructed models consisting of the combination of the input variables and the best fit input structure was investigated. The performance of the constructed ANNs models in training and testing processes were compared with the measured soil temperature values to identify the best fit forecasting ANN model. Our results showed that the Nonlinear Autoregressive ANN approach are best model for forecasting the soil temperature of the day.

Read Publication

Backpropagation Neural Network Algorithm for Forecasting Soil Temperatures Considering Many Aspects: A Comparison of Different Approaches

Journal ICIT 2011 The 5th International Conference on Information Technology

Artificial Neural Networks (ANNs) are interconnected collections of processing units which have been used in different applications. The objective of this paper is to develop an ANN model to estimate soil temperature for any day by using various previous day meteorological variables. For this purpose, average temperature of air, sunshine, radiation and soil temperature for meteorological data between the years of 1980 and 1984 at Nineveh/Iraq Meteorological Station were used. We measured the soil temperatures at different depths of 5, 10, 20, 50 and 100cm within the time 9, 12 and 15 respectively. Three ANNs models were constructed. The Backpropagation neural network algorithm (BP), Cascade-Forward and Time Series (or Nonlinear Autoregressive) algorithms were used for the training the constructed ANNs models. These constructed models consisting of the combination of the input variables and the best fit input structure was investigated. The performance of the constructed ANNs models in training and testing processes were compared with the measured soil temperature values to identify the best fit forecasting ANN model. Our results showed that the Nonlinear Autoregressive ANN approach are best model for forecasting the soil temperature of the day.

Read Publication

NONLINEAR AUTOREGRESSIVE NEURAL NETWORK FOR ESTIMATION SOIL TEMPERATURE: A COMPARISON OF DIFFERENT OPTIMIZATION NEURAL NETWORK ALGORITHMS

Journal UbiCC Journal

Issue Special Issue of ICIT 2011 Conference

The objective of this paper is to develop an Artificial Neural Network (ANN) model to estimate soil temperature for any day. We used air average temperature, sunshine, radiation and soil temperature for meteorological data between years [1980 and 1984] at Nineveh/Iraq Meteorological Station. In this research, three ANN models with their associated training algorithms (Backpropagation neural network (BPNN), Cascade-Forward and Nonlinear Autoregressive (NARX)) were used for estimating soil temperatures at different depths of 5, 10, 20, 50 and 100cm within the time 9, 12 and 15 respectively. The performance of the three models with their training algorithms were compared with the measured soil temperature values to identify the best fit ANN model for soil temperature forecasting. The results showed that the NARX model is the best model. Finally, a comparison between five optimization ANN training algorithms was adopted to train NARX ANN model to identify best fit optimization algorithm for forecasting soil temperature with best results. From comparisons, TrainLM is the best optimization algorithm for training NARX model.

تمييز أصوات الأرقـــام العــربيـة

Journal مجلة تكريت للعلوم الصرفة

publisher جامعة تكريت

Volume Volume 12

Read Publication