مجله رایانش نرم و فناوری اطلاعات
سال چهارم شماره 3 (پاییز 1394)

In this paper, a new image watermarking method in Discrete Fourier Transform (DCT) domain based on Fuzzy logic and human vision system is proposed. In first stage, input image is divided into 8*8 blocks and then the Fuzzy system allocates a coefficient for each block based on texture features and illumination which results in more robustness and clarification in watermark process. Next, by calculation of Discrete Fourier Transform (DFT) for each block, the resulting magnitude and phase are separated from each other and watermark is hidden in magnitude. The process of watermark extraction is blind which corresponds to higher security. The shortcoming of the general watermark system DCT-based is that the human vision system is neglected and therefore in order to prevent degradation of image quality, less robustness is resulted. The experimental results show that the proposed method provides more illumination and more robustness against cutting attacks, brightness and darkness of image, additive noise, image filtering and compression in comparison with popular DCT-based methods.

Travel time prediction has been concerned by specialists in different fields in regard with transportation systems programming. Geospatial information systems and data mining analyzes by using new methods have obtained several capabilities to extract hidden patterns among the data to predict travel time. These methods can be considered as good alternatives for conventional methods. In this study K-means clustering method as a spatial data mining technique and also a forward neural network method have been used to predict travel time. In this regard eight variables were extracted from vehicles GPS and climatic and cars accidents data during a three years period. These variables were used to predict travel time in Babol – Ghaemshahr and Ghaemshahr – Sari roads in Mazandaran province. In order to evaluate the proposed method the outputs of clustering and neural network methods were compared with real data in the study corridor. The study showed the artificial neural network method can lead to more satisfactory results in comparison with the clustering method by considering a root mean square error equal to 0.0015 and a mean square error equal to 0.65 showed.

The Delay Tolerant Mobile Sensor Networks (DTMSNs) distinguish themselves from conventional sensor networks by means of several features such as node mobility, loose connectivity, and delay tolerability. In comparison with conventional networks, prominent feature of these networks is that there is not end-to-end path between source and destination. Due to energy limitation, one of the crucial issues in large-scale networks is energy storage that can be overcome this problem by distribution of multi-sink in a network at the same time. In this paper, in order to covering large area, we determine a new radio range to sink nodes that prevents the presence of nodes in the radio range of multiple sinks and sending data to a large number of sinks. Then, we represent Replica Adaptive Data Gathering Protocol. Simulation results indicate that the proposed protocol significantly improves message delivery rate with lower transmission overhead and also lower delay in data delivery than the other existing DTMSNs routing protocols.

Mobile ad hoc networks consist of mobile nodes without any infrastructure. Movement of nodes in mobile ad hoc networks is free in each side. These networks are confronted with several limitations that should be considered in the design and implementation of any protocol. Due to the dynamicity of these networks, supporting Quality-of-Service is a primary requirement of the applications deployed over MANET‟s. To service based on quality of service, we have proposed a new protocol that search and allocate services between applicant nodes in the network. The main feature of this algorithm, is selecting servers based on the current state of the network that can reduce the number of service handoffs and serve the clients according to their performance needs, specified in the client Service Level Argument. An emulator has been used to evaluate the proposed protocol. It was observed that applying this algorithm has significant impact on the quality of service parameters.

Steganography is a branch of covert communication science, that the purpose is to hide the existence of information.Versus,the purpose of steganalysis is to detect and/or estimate potentially hidden information from observed data with little or no knowledge about the steganographic algorithm or its parameters. Although, the color images is common and have high capacity steganography, but researches in hiding information in color images is less than grayscale images. In this article, we have studied steganography and steganalysis methods in color images using color space information (such as RGB, YUV, YIQ, YCbCr and HSV color space) and proposed new methods in two domains. In steganography domain, we have proposed a simple and robust method for hiding information in the YUV and YCbCr color images using color spaces, that recognition of the message has become more difficult due to the use of conversion coefficients. The proposed method is evaluated by several famous steganalysis, and the results show that resistance of this method enhanced in comparison RGB color space. In steganalysis domain,we have examining the different color spaces and we have proposed a new general steganalysis method that based on spatial correlation of adjacent pixels in different color spaces components. It is independent of the type of steganography designed and has good detection. In general, the results of the two methods is highlights the effective and appropriate use of hiding information in color images in color space.

In this paper Behavioral scoring model that combined analytic hierarchy process and data mining techniques is proposed to assessment Credit risk of real customers and providing organization knowledge that are helpful to decide whether or not to grant amenities to applicants . Therefore, in step one an improved data preparation method was applied to prepare and select input feuteres of behavioral scoring model. in this step we tried to extract new features that covers through interaction between customers and bank. Then analytic hierarchy process (AHP) was applied to determine the relative weights of customer’s behavioral predicators. in modeling phase, behavioral scoring of customers was defined Base on their repayment behavior and late repayments duration. The dataset that has used in this article is provided by an Iranian private bank. The proposed model show that better result and also demonstrate that the hybrid mining approach and AHP can be used to build effective behavioral scoring models better result. Therefore, finance and banking institution can utilize the novel model to identify and predict customer’s credit behavior and decrease credit risks.

In this paper, feature fusion technique is employed for improvement of recognition of handwritten digits. By merging three different feature vectors, given a specific weight for each of vectors, the Genetic Algorithm and Particle Swarm Optimization processes were applied to calculate the optimum weights. The main objective in this study was to compare the calculated weights according to each of the optimization techniques to that of classifiers combination in order to achieve a higher recognition rate and time for Persian Handwritten digits. A database containing 60'000 training samples and 20'000 test samples is used for the process.

Nowadays the artificial intelligent algorithms has considered in GIS and remote sensing. Genetic algorithm and artificial neural network are two intelligent methods that are used for optimizing of image processing programs such as edge extraction and etc. these algorithms are very useful for solving of complex program. In this paper, the ability and application of genetic algorithm and artificial neural network in geospatial production process like geometric modeling of satellite images for ortho photo generation and height interpolation in raster Digital Terrain Model production process is discussed. In first, the geometric potential of Ikonos-2 and Worldview-2 with rational functions, 2D & 3D polynomials were tested. Also comprehensive experiments have been carried out to evaluate the viability of the genetic algorithm for optimization of rational function, 2D & 3D polynomials. Considering the quality of Ground Control Points, the accuracy (RMSE) with genetic algorithm and 3D polynomials method for Ikonos-2 Geo image was 0.508 pixel sizes and the accuracy (RMSE) with GA algorithm and rational function method for Worldview-2 image was 0.930 pixel sizes. For more another optimization artificial intelligent methods, neural networks were used. With the use of perception network in Worldview-2 image, a result of 0.84 pixel sizes with 4 neurons in middle layer was gained. The final conclusion was that with artificial intelligent algorithms it is possible to optimize the existing models and have better results than usual ones. Finally the artificial intelligence methods, like genetic algorithms as well as neural networks, were examined on sample data for optimizing interpolation and for generating Digital Terrain Models. The results then were compared with existing conventional methods and it appeared that these methods have a high capacity in heights interpolation and that using these networks for interpolating and optimizing the weighting methods based on inverse distance leads to a high accurate estimation of heights.

In recent years, GPS has attracted the attention of many users in the industrial, military and commercial fields due to its accurate time and position information. Because the received signal power on the earth surface is lower than the thermal noise level, it can seriously subject to intentional or unintentional interferences. Intentional interference is known as ‘jamming’. Although the GPS spread-spectrum signal structure has some inherent jam protection, when a hostile jammer want to disturb a GPS system need only send out a jamming signal with enough power and suitable time/frequency properties to deny the use of GPS. Using Neural Networks (NNs) is a non linear filtering approach for tracking and canceling interference. In this paper, we investigate one of the NNs structures (multi-layer perceptron) and the possibility of interference elimination using this network. Finally, the proposed method will be compared with one of wavelet structures. It can be seen that the proposed algorithm identifies more than four satellites for solving the navigation equations. In addition, it is robust against the increment of jammer power ( from 25dB to 50dB) and improves the similarity of predicted signal to the real one about 45% in comparison with the wavelet structure.

The information extraction of urban features is important for managers and city planners. In this way, we can use remote sensing data and related methods in order to detection of these features efficiently. In this study, Object-based and pixel-based detection of urban features is done by integration of LIDAR data and large-scale aerial optical images at the level of decision. Pixelbased and object-based analysis is done based on decision tree classification with consideration of shadow and without it. This study demonstrates ability of LIDAR data to solve problems caused by the shadow in urban area. The accuracy of object based method is more than pixel based method in two classifications. In object based analysis over all accuracy in both classifications are similar, but classification with shadow class is better than other method. Over all accuracy of pixel based classification without shadow class is 0.91 which is the highest accuracy in pixel-based analysis.

In this paper, a novel method is proposed to detect and to track pupil of one eye using modified circular Hough transform and a pair of goggles equipped with an infra -red (IR EMD) camera. The IR-EMD, which is used in this research, is designed and built after studying several different schemes and despite its simplicity, it has very good performance and could be thought as one the practical novelties of the project. In the proposed method for pupil detection, circular Hough transform is used and its speed and accuracy is enhanced performing several modifications. Applying the modified circular Hough transform to the image captured with an average quality camera of the IR-EMD, which is also equipped with an IR-LED ring, the accurate position of the pupil would be detected. In addition to high speed and precision, the algorithm does not require high quality camera and is robust to the input image noise and some other environmental parameters such as rotation of the head, eye make-up and background light of the image, as well. Practical experiments on several people, indicate precision and performance of the proposed algorithm. Besides, in order to confirm the performance of the algorithm for almost all possible situations, the method is applied to more than 700 eye images of the CASIA-Iris V.3 database. Results show outstanding performance of the proposed method. For instance, in presence of pepper and salt noise with 50 percent intensity, the algorithm could detect the correct position of the pupil of the database images with almost 90 percent accuracy.

In this paper, a new model for mobile robot mapping is presented. The model is inspired by functionality of the cells of cortex postsubiculum layer. This model is based on visual information so, robot visual input from environment is considered as input of the model. 1 layer of visual cortex of the brain, is modelled by Gabor filter due to extracting image texture and Gabor filter histogram is used as image features. Therefore, the model can be used in real environments with similar colors. The output dimension of this layer is decreased using unsupervised basic dimension reduction techniques such as Kernel-PCA, PCA, ISOMAP and MDS. High-dimensional data suffer from problems called curse of dimensionality. By reducing the data dimension, in addition to the reduceddata volume storage, this problem is to overcome. To the best of our knowledge, the model is the first model that was developed with the purpose of dimension reduction. Another innovation is presentinga fuzzy clustering model. Using limited number of direction cells, the model makes interpolate possible to find robot head direction in defuzzification step. In previous models such as Tokonaga and Milford, output angels are limited to number of direction cells, while this constraint is resolved in the proposed model. The output of direction cells provided by the model are similar to the actual output of the direction cells that have been obtained from experimental tests on the brain. The implementation results of the proposed model is evaluated and compared with other methods. In most cases results show higher accuracy.

Economic dispatch at minimum production cost is one of the most important subjects in the power system operation, which is a complicated nonlinear constrained optimization problem. To solve the dynamic economic dispatch problem, it is assumed that a thermal unit commitment has been previously determined. Since dynamic economic load dispatch was introduced, several methods have been used to solve this problem. However, all of those methods may not be able to provide an optimal solution and usually getting stuck at a local optimal. In this paper, an Improved Particle Swarm Optimizer (IPSO) has been proposed to solve dynamic economic load dispatch problem. This algorithm is applied to a complex dynamic economic dispatch problem for 6-unit power systems with a 24-h load demand at each 1-h time intervals. Comparing IPSO results with other method's results that reported in literature shows the ability of IPSO to reach better solutions.