(0 Item)

Proceedings of the Advanced Topics in Artificial Intelligence (ATAI 2010)

Academic Conferences - Proceedings of the Advanced Topics in Artificial Intelligence (ATAI 2010)
Academic Conferences

By : Global Science & Technology Forum

Date : 2010

Location : Thailand / Phuket

PDF 117p
Description :

The ATAI conference presented a mix of theory and applications on a wide spectrum of advanced topics of artificial Intelligences such as Neural Networks,Image Segmentation or Knowledge Reasoning

Keywords :

Image Segmentation, Neural Networks, Pulse Coupled Neural Network, Two-layer PCNN, Image Processing, Image Interpretation, Curve Matching, Rotation Independent Hierarchical , machine learning, Boundary Representation, fingertips detection, and tracking, Obstructive Sleep Apnea (OSA), Wavelet Packets, Spectral Analysis, Artificial Neural networks (ANN), cross-correlatio

Keywords inside documents :

image ,curve ,memory ,based ,network ,domain ,click ,model ,query ,linking ,layer ,pleasure ,system ,method ,algorithm ,neurons ,resources ,object ,objective ,agent

Proceedings from Advanced Topics in Artificial Inelligence (ATAI 2010) conference.

 

 

Over decades, significant advances have been made in a number of core problems of artificial intelligence (AI).  The ATAI conference will feature a mix of theory and applications on a wide spectrum of advanced topics of AI. On top of core areas such as machine learning, neural networks, and soft computing, the concept of AI has been applied to problems in other areas of computer science as well (e.g., computer networks, intelligent databases, information retrieval, program synthesis, automated discovery, automated design, robotics, operating systems, parallel and distributed computing).

 

 

 

 

Abstracts of papers included in the Proceedings from Advanced Topics in Artificial Inelligence (ATAI 2010) conference. 
 
A BAYESIAN FRAMEWORK FOR FEEDBACK IN REGION BASED IMAGE RETRIEVAL

Manjeet Rege, Department of Computer Science, Rochester Institute of Technology, Rochester, NY, USA & Rajeev Agrawal, Department of Electronics, Computer & Info. Tech. North Carolina A&T State University, Greensboro, NC, USA.

 

Relevance feedback has been long used as a tool for improving the performance of image retrieval. We present a Bayesian framework that captures and synthesizes the user feedback at an object or image region level. By exploiting the statistical structure of images, our system is able to discover the object of user interest. First, all the images in the database are segmented and the image regions are clustered into different region clusters. Next, for every region in the query image, we find the representative cluster that has the highest posterior probability, given the image region. As feedback is received, the cluster priors change, leading to different clusters competing for a image region. We have integrated our region based feedback mechanism into a image retrieval system. Preliminary experiments performed on general purpose images demonstrates the promise of the proposed framework.

 

INTELLIGENT KALMAN FILTER MODEL FOR TARGET TRACKING

Prof A. Surendra Rao, (Former Scientist-F, Additional. Director, DRDO, MOD) Computer Science & Engg,

Raghu Engineering, College Dakamarri, Bheemili, Visakhapatnam,AP,India  & RV Kiran Kumar, Scientist-B,CSS Division, NSTL, Visakhapatnam ,AP, India.

 

The kalman filter has been used in the state estimation of target in question. But during target manoeuvre filter performance is seriously degraded because target dynamics appears as extensive noise on the object model. A new intelligent kalman filter (IKF) is

derived for tracking the manoeuvring target model in which (the unknown target acceleration is regarded as additive process noise) the time varying variance of process noise is computed in an intelligent manner using a fuzzy system. To optimize the fuzzy system, a genetic algorithm (GA) and DNA coding methods are utilized. Different target models are applied to IKF and target parameters are computed and presented (range, bearing, velocity and course). The faster convergences of target parameters are achieved in the model. The GA and DNA coding techniques are compared and results are plotted. The optimization of fuzzy logic with DNA shows better results than fuzzy logic with GA. The results are presented for different target models.

 

Two-Layer Recurrent Pulse Coupled Neural Network for Image Segmentation

Heggere S. Ranganath and Ayesha Bhatnagar from The University of Alabama in Huntsville, Huntsville, Alabama, 35899, USA.

 

For over a decade, the Pulse Coupled Neural Network (PCNN) based algorithms have been used for image segmentation. Though there are several versions of the PCNN based image segmentation methods, almost all of them use single-layer PCNN with excitatory linking inputs. Often the PCNN parameters including the linking coefficient are determined by trial and error. This paper presents a new 2-layer network organization for PCNN in which excitatory and inhibitory linking inputs exist. The value of the linking coefficient and the threshold signal at which primary firing of neurons start are determined directly from the image histogram. Simulation results show that the new PCNN achieves significant improvement in the segmentation accuracy over other methods including the widely known Kuntimad’s single burst image segmentation approach. The improvement is due to the fact that neurons corresponding to spatially adjacent regions compete to capture neurons corresponding to boundary pixels. Simulation results also show that small or even moderate increase in the value of the linking coefficient from its optimal value has practically no adverse impact on the segmentation accuracy.

 

 

Rotation independent hierarchical representation for Open and Closed Curves

Siddharth Shivapuja1, Vineetha Bettaiah2, Thejaswi Raya3 andHeggere Ranganath4

1Honeywell Scanning and Mobility, Blackwood, NJ 08012, USA; 2The University of Alabama in Huntsville, Huntsville, AL 35899, USA; 3The University of Alabama in Huntsville, Huntsville, AL 35899, USA; 4The University of Alabama in Huntsville

Huntsville, AL 35899, USA.

 

The algorithm used for the segmentation of an image, and scheme used for the representation of the segmentation result are mostly selected based on the final image analysis or interpretation objective. The boundary based image segmentation and representation system developed by Nabors segments and stores the result as a graph-tree hierarchical structure without any kind of prior knowledge of the final image analysis or interpretation objective [1]. The representation allows the development of efficient feature extraction and interpretation algorithms to support diverse image processing applications such as object recognition, scene matching, content based image retrieval, etc. This paper shows that Nabors’ hierarchical representation of curves

is not invariant to rotation, and proposes an enhanced representation which retains its structure and remains invariant under rotation. The new representation makes it easy to determine if a curve is a section of a larger curve.

 

Hand Feature Detection from Skin Color Model with Complex Background

Ahmad Yahya Dawod, Faculty of Information Technology, Multimedia University, Cyberjaya, Malaysia; Junaidi Abdullah, Faculty of Information Technology, Multimedia University, Cyberjaya, Malaysia. Md. Jahangir Alam, Faculty of Information Technology, Multimedia University, Cyberjaya, Malaysia

 

We present a new technique for extraction of hand region from complex background and consequently detection of the fingertips, which we call hand features, from color images. Our construction is primarily based on an adaptive color model generation for hand segmentation followed by smoothing algorithm. We present a thinning algorithm followed by the construction of convex envelope to detect possible points for the fingertips. We demonstrate that the correct points for the fingertips can be selected heuristically through interpoints distance calculation. Finally, we show the effectiveness of the proposed method by experimenting with images of different background complexity and have achieved very promising results.

 

Partition Fusion Approach of Image Fusion Using Modified Pulse Coupled Neural Network for Differently Focused Images

Dheeraj Agrawal, Electronics and Communication Engineering Department, Maulana Azad National Institute of Technology

Bhopal, India; Jyoti Singhai, Electronics and Communication Engineering Department, Maulana Azad National Institute of Technology,Bhopal, India.

 

The paper introduces an approach for fusion of differently focused images for improved human perception and machine recognition. The fusion algorithm uses a partition fusion approach where source images are Partioned into blocks and based on their clarity these blocks are selected for composite image. The quality of composite image thus depends upon the clarity measure and decision making system to select a block from given multi focus images. In the proposed algorithm the clarity of the image blocks is decided by EOL. The selection of blocks is performed by the Modified PCNN based on the clarity of the image blocks. To improve the performance of PCNN some modification are used in feeding and linking filed in PCNN. The performance of the proposed fusion algorithm has been tested for RMSE, Mutual Information and standard deviation. Considerable improvements are observed in the results as compared to existing method of image fusion.

 

Identification of Patients with Obstructive Sleep Apnea Using Wavelets Packets and Artificial Neural Networks

Abdulnasir Hossen, Member IEEE, Department of Electrical and Computer Engineering, Sultan Qaboos University

PO Box 33, PC 123, Al-Khoud, Sultanate of Oman.

 

A new identification method for identification of patients with obstructive sleep apnea (OSA) from normal controls is investigated in this paper using estimated spectral analysis of RRI data with wavelets packets and artificial neural networks. Two sets of data are used in this paper. The training data is obtained from Sultan Qaboos University

hospital while the test data is obtained from MIT databases. The training data set consists of 15 OSA and 15 normal subjects. The test data set is divided into two test sets each consists of 20 OSA and 10 normal subjects. The spectral analysis of RRI data obtained using 8 different sub-bands from wavelets packets is used as a classification feature. A simple artificial neural network of the type feed-forward back-propagation is used for the classification task. Different types of wavelets are used to test the consistency of the approach. The accuracy of classification approaches 92.7% using a large size of data simulated with power spectral density values of the main 8 sub-bands within the mean value plus/minus the standard deviation of the power spectral density values of the original test data sets.

 

Fast Fractal Encoding through FFT using Modified  Crosscorrelation based Similarity Measure

S.B Dhok,R.B.Deshmukh,A.G. Keskar, Visvesvaraya National Institute of Technology,Nagpur (INDIA).

 

The image compression using fractal transform is a promising method which is potentially capable of achieving very high compression ratios. The major drawback of fractal inage compression is large encoding time, though the decoding time is negligible. In this paper, a new similarity measure based on normalized cross-correlation of mean subtracted range and domain blocks is proposed.The fast fractal encoding algorithm based on the proposed similarity measure is well suited for FFT based frequency domain operations to speed up the encoding process. The implemented algorithm employs exhaustive search of similar domain blocks for each range block unlike other limited domain search methods. The algorithm works largely in frequency domain and operates on entire domain image instead of overlapping domain blocks.The contrast and brightness parameters of fractal transformation are easily calculated during the course of computation of similarity index matrix. Though the proposed method shows little dB drop in Peak signal to noise ratio(PSNR) values, the encoding time is reduced considerably with average speedup factor of 30 as compared to the full search method.

 

Intelligent Resource Selection for Sensor-Task Assignment: A Knowledge Based Approach

Geeth de Mel, Wamberto Vasconcelos, Timothy J. Norman, Department of Computing Science, University of Aberdeen, Aberdeen AB24 3UE, Scotland, United Kingdom

 

Sensing resources play a crucial role in the success of critical tasks such as surveillance. Therefore, it is important to assigning appropriate sensing resources to tasks such that the selected resources fully cater the needs of the tasks. However, selecting the right resources to tasks is a computationally hard problem to solve. Most of the existing approaches address the efficiency aspect of the resource selection by considering the physical aspects of the sensor network (e.g., range, power, etc.) but have ignored important domain related properties such as capabilities of assets, environmental conditions, policies and so on which makes the selection effective. In this paper we present a knowledge rich mechanism to intelligently select ressources for tasks such that the selected resources sufficiently cover the needs of the tasks. Ontologies are used to capture the crucial domain knowledge and semantic matchmaking is used to perform sensor-task matching. A combination of ontological and first-order-logic reasoning is considered for the solution architecture.

 

Multi-click dependent model to estimate document relevance in web search

Xinyi Shu , Yujiu Yang†, Wenhuang Liu, Graduate School at Shenzhen, Tsinghua University, Beijing, China;

†Graduate School at Shenzhen, Tsinghua University, Beijing, China; ‡Graduate School at Shenzhen, Tsinghua University, Beijing, China.

 

Web search click logs, reflecting whether users are satisfied with the search results, are the most extensive and invaluable information resources of user preference. A central problem in click log analysis is to estimate the userperceived relevance of each query-URL pair. Many click models have been proposed to solve this problem, but they all have a common problem: the examine-next probability only depends on the current result or the preceding last clicked result. Intuitively, whether a user will continue to see the next result is supposed to be determined by whether one is satisfied with the information got from one’s historical clicked results, not only the current result. Therefore, we propose the multi-click dependent model (MCDM) that takes all the preceding clicked results into consideration. In the new model, the examine-next probability is decided by the click variables of each clicked result. We evaluate the proposed model on a real-world data set consisting of about 3.02 million query sessions obtained from a Chinese commercial search engine Sougou to test the performance of MCDM. The experiment results show that MCDM outperforms the existing click models in metrics such as log-likelihood, click perplexity, last click prediction error, especially on less-frequent queries and bottom positions of query sessions.

 

On Fuzziness in Hybrid Network Device Discovery

Mustafa Abdat , Pradeep Isawasan , Muhammad Fermi Pasha†, Ahmed M. Manasrah,  National Advanced IPv6 Centre, Universiti Sains Malaysia, 11800, Minden, Penang, Malaysia. †School of Computer Sciences, Universiti Sains Malaysia, 11800, Minden, Penang, Malaysia.

 

The importance of network discovery cannot be denied especially for network monitoring and network management purposes. Here, we present a fuzzy approach to perform hybrid network device discovery. The main goal is to solve the increased additional traffic load issues in active discovery technique and the low accuracy issues in passive discovery technique. Then to further otptimize the hybridization between the passive and active approach, a fuzzy values are set to alternate the use of active and passive technique efficiently. Our preliminary results show that the proposed work are able to discover devices within the network-in-test with less additional traffic but with high accuracy.

 

NEURO-SYMBOLIC INTEGRATION USING PSEUDO INVERSE RULE

Saratha Sathasivam & 1Muraly Velavan; School of Mathematical Sciences, Universiti Sains Malaysia, 11800 USM, Penang; 1 Surya College, 12200 Butterworth, Penang

 

Pseudo inverse learning rule is examined for its ability to accelerate the performance of doing logic programming in Hopfield neural network. This learning rule has a higher capacity than the Hebb Rule and Direct learning rule. This learning rule also suffers significantly less capacity loss as the network gets larger and more complex. Comparisons are made between these three rules to see which rule is better or outperformed other rules in the aspects of computation time, memory and complexity.

 

Identifying People’s Intention from Natural Language Texts

Sirichai Triamlumlerd & Dr.Jeremy Ellman

 

This research is aimed at analyzing intention from ordinary language texts and producing a computer model of human agents, their intentions and actions. It will focus

on words such as ‘assault’ which is the semantic key to recognize the people’s intentions. The input texts to analyze intention are from online legal databases. To ensure that the model of intention analysis is language independent, it will be developed and tested in languages from two different linguistic groups, English and Thai. An approach is proposed that merges machine learning with ontological analysis

 

Generation of Anaphoric Noun Phrases and their coreferential properties

Parma Nand, School of Computing and Mathematical Sciences, Auckland University of Technology, Auckland, New Zealand & Wai Yeap, School of Computing and Mathematical Sciences, Auckland University of Technology, Auckland, New Zealand.

 

This paper investigates the generative processes involved in the formation of noun phrases (NPs) from an existing clause in a discourse. A speaker or writer is able to generate a range of NPs from a stated clause to co-refer to the different semantic units of the clause by combining the various primitive components. These generative processes can be grouped into either predicate-deletion or normalization. We use relations defined by Levi [1] as a basis to define constraints used in the formation of NPs by predicate-deletion and empirically determine the conformance level of these constraints with natural discourses. For the normalization process, we argue that the inflicted morpheme’s anaphoric properties can be characterized by their suffixes. We also show that, these anaphoric properties are independent of the choice between subjective and objective modifiers. The anaphoric properties of suffix-based morphemes is also empirically investigated for conformance levels, and the results are encouraging.

 

Factors that Influence the Memory of Entertainment Experiences

Shenglan Kang and Alexander Nareyek

 

When generating an entertainment experience, such as that of a movie, a novel, or a computer game, not only the experience itself, but also the memory of the experience plays an important role. Pleasurable memories can generate additional pleasure during recall, be instrumental to generate secondary social pleasure when communicating the experience, and be important for word-of-mouth advertising of the entertainment product. In order to optimize entertainment experiences, especially for an online dynamic generation of experiences within computer games, we need to build formal models of these processes. This paper is a summary of a research study we conducted with 50 university students in order to identify factors that influence the memory of entertainment experiences. Our findings show that when watching a short film, a range of factors, including the level of attention of the person, the predictability of the storyline, and the valence of the content, all have substantial impact on the amount of information the test subject can recollect at a later stage. Psychophysiological measurements like skin conductance are also found to be indicators of the memory performance. In terms of the application of the results, some of the factors suggest the use of profiling techniques to best match the targtet audience's background, e.g., when used for automated story generation within a game.

 

Using Hash Tables to Expedite Knowledge Reasoning in the General Game Playing Agent

Xinxin Sheng, David Thuente, Computer Science Department, North Carolina State University, Raleigh, USA.

 

General game playing research focuses on designing automated agents that accept declarative logic description of arbitrary games at run time and are able to play efficiently without human intervention. The game information including the game states, the rules of the game, and the player's role in the game are all represented in logic relations. The general game playing agent uses knowledge representation and reasoning algorithms to analyze and play the game. We use hash table to significantly improve the reasoning performance. We provide experimental data on seven different games: small games like the single player game Maze, the strategy game Mini-Chess, the two player game Tic-Tac-Toe, the middle size board games Connect Four and Chess Endgame, the large size game like Othello, and finally the three-player eCommerce game Farmer. In all of these scenarios, our agent has proven to significantly outperform the standard published Java player

 

Use of Probabilistic State Diagrams for Robot Navigation in an Indoor Environment

Bogdan Czejdo and Sambit Bhattacharya, Department of Mathematics and Computer Science, Fayetteville State University, Fayetteville, NC 28301, & Janusz Czejdo, Department of Foreign Languages, Edinboro University, Edinboro, PA 16444

 

This paper describes the analysis of syntax and semantics of state diagrams to support probabilistic behavior of robots. As a result we created probabilistic state diagrams that can be used for automatic code generation. The integration of the advanced robot vision algorithms with probabilistic state diagrams is also addressed. It is shown how to use techniques such as state abstraction and transition abstraction to create, verify and combine large probabilistic state diagrams. The paper also describes the implementation model for probabilistic state diagrams.

 

Pleasure Differences as a Result of Seeing an Action versus Own Acting

Xinying Cheah and Alexander Nareyek

 

Actively playing a computer game will generate more pleasure than a corresponding passive experience, such as watching a movie or reading a book. This is what common knowledge tells us. It is however not fully consistent with our research on the human reward system, and we undertook the study described in this paper to have a closer look at the differences. In this study, test participants watched videos of actors playing games as well as actively played the games themselves. The study examined how the pleasure difference of the participant's own acting versus pasively warching the  actions of an actor changes with the participant's level of empathy, the sympathy toward the actor, and the level of emotion displayed by the actors. Among other results, the findings indicate that a higher pleasure resulting from own actions diminishes with a higher empathy rating, with equal pleasure at about the maximum rating of 80.

 

PPNMF: Improving Weighted Nonnegative Matrix Factorization with Prior Information

Shuli Han 1, Yujiu Yang 2, Wenhuang Liu 3, 1,2,3Graduate School at Shenzhen, Tsinghua Universtity

Shenzhen 518055, P.R.China.

 

Collaborative Filtering (CF) is one of the most popular methods for recommendation problem. The key idea is to predict further the interests of a user (ratings) based on the

available rating information from many users. Recently, matrix factorization(MF) based approaches, one branch of collaborative filtering, have proven successful for the rating prediction issues. As is well known, most of the MF models follow one of the following frames: (1) to fit a linear factor model over all observed ratings respect to the Frobenius norm with a positive regularization item, (2) to fit a linear factor model over all observed ratings respect to the Frobenius norm with a nonnegative constraint added. Differing from the exiting MF models, which model only on the observed ratings, the proposed variant of MF, referred to as Probabilistic Prior Nonnegative Matrix Factorization (PPNMF) in this paper, utilizes the prior information of the missing elements via treating each missing as a random variable; the probability distribution of each element is calculated with some specific scheme. Compared with the traditional Matrix Factorization for Collaborative Filtering, our empirical studies show that the proposed algorithm makes more accurate predictions of user ratings and is more robust with respect to the initial setting. With this method, we analyze the mechanism of its learning process, which shows that the algorithm will first go through a damping vibration process to make an adjustment, and then converges.

 

Solving Multi-Objective Problems under Its Objective Bounds by Genetic Algorithms

Anon Sukstrienwong, Department of Information Technology, Bangkok University, Bangkok, Thailand

 

In this paper, a new approach is presented to solve the particular problems in which some objective functions are controlled to be within its objective bounds. The proposed algorithm called GABound algorithm is based on genetic algorithms (GAs) for searching the multi-objective solutions. The algorithm employs the preemptive optimization technique by considering one objective at a time based on the priority of the objective functions. The simulation results indicate that the proposed algorithm is satisfactory with customization of the number of eras and immigration rate.

 

PID-GPSO Load Frequency Controller for Interconnected Power System

Amer S. Al-Hinai, Sultan Qaboos University, P.O. Box 33 Al-Khodh, Muscat 123; Sultanate of Oman

 

This paper is proposing a Load Frequency (LF) controller design for interconnected power system based on Proportional Integral Differential (PID) controller. The PID gains have been design using Guided Particle Swarm Optimization (GPSO) technique. The controller is designed to improve the dynamic response of system frequency and the tie line power flow under a sudden load changes. The LFC model for a two area interconnected power system is implemented in SIMULINK/MATLAB and the simulation results illustrate the effectiveness of the proposed LF controller compared to integral controller.

A Bayesian Framework for feedback in region based Image retrieval

Company Description : Relevance feedback has been long used as a tool for improving the performance of image retrieval. We present a Bayesian framework that captures and synthesizes the user feedback at an object or image region level. By exploiting the statistical structure of images, our system is able to discover the object of user interest. First, all the images in the database are segmented and the image regions are clustered into different region clusters. Next, for every region in the query image, we find the representative cluster that has the highest posterior probability, given the image region. As feedback is received, the cluster priors change, leading to different clusters competing for a image region. We have integrated our region based feedback mechanism into a image retrieval system. Preliminary experiments performed on general purpose images demonstrates the promise of the proposed framework.

Product Type : Academic Conferences

Author : Manjeet Rege, Department of Computer Science, Rochester Institut

PDF 2p

Languages : English

Intelligent Kalman filter model for target tracking

Company Description : The kalman filter has been used in the state estimation of target in question. But during target manoeuvre filter performance is seriously degraded because target dynamics appears as extensive noise on the object model. A new intelligent kalman filter (IKF) is derived for tracking the manoeuvring target model in which (the unknown target acceleration is regarded as additive process noise) the time varying variance of process noise is computed in an intelligent manner using a fuzzy system. To optimize the fuzzy system, a genetic algorithm (GA) and DNA coding methods are utilized. Different target models are applied to IKF and target parameters are computed and presented (range, bearing, velocity and course). The faster convergences of target parameters are achieved in the model. The GA and DNA coding techniques are compared and results are plotted. The optimization of fuzzy logic with DNA shows better results than fuzzy logic with GA. The results are presented for different target models.

Product Type : Academic Conferences

Author : Prof A. Surendra Rao, (Former Scientist-F, Additional. Director,

PDF 6p

Languages : English

Two-Layer Recurrent Pulse Coupled Neural Network for Image Segmentation

Company Description : For over a decade, the Pulse Coupled Neural Network (PCNN) based algorithms have been used for image segmentation. Though there are several versions of the PCNN based image segmentation methods, almost all of them use single-layer PCNN with excitatory linking inputs. Often the PCNN parameters including the linking coefficient are determined by trial and error. This paper presents a new 2-layer network organization for PCNN in which excitatory and inhibitory linking inputs exist. The value of the linking coefficient and the threshold signal at which primary firing of neurons start are determined directly from the image histogram. Simulation results show that the new PCNN achieves significant improvement in the segmentation accuracy over other methods including the widely known Kuntimad’s single burst image segmentation approach. The improvement is due to the fact that neurons corresponding to spatially adjacent regions compete to capture neurons corresponding to boundary pixels. Simulation results also show that small or even moderate increase in the value of the linking coefficient from its optimal value has practically no adverse impact on the segmentation accuracy.

Product Type : Academic Conferences

Author : Heggere S. Ranganath and Ayesha Bhatnagar from The University of

PDF 6p

Languages : English

Rotation independent hierarchical representation for Open and Closed Curves

Company Description : The algorithm used for the segmentation of an image, and scheme used for the representation of the segmentation result are mostly selected based on the final image analysis or interpretation objective. The boundary based image segmentation and representation system developed by Nabors segments and stores the result as a graph-tree hierarchical structure without any kind of prior knowledge of the final image analysis or interpretation objective [1]. The representation allows the development of efficient feature extraction and interpretation algorithms to support diverse image processing applications such as object recognition, scene matching, content based image retrieval, etc. This paper shows that Nabors’ hierarchical representation of curvesis not invariant to rotation, and proposes an enhanced representation which retains its structure and remains invariant under rotation. The new representation makes it easy to determine if a curve is a section of a larger curve.

Product Type : Academic Conferences

Author : Siddharth Shivapuja1, Vineetha Bettaiah2, Thejaswi Raya3 andHegg

PDF 6p

Languages : English

Hand Feature Detection from Skin Color Model with Complex Background

Company Description : We present a new technique for extraction of hand region from complex background and consequently detection of the fingertips, which we call hand features, from color images. Our construction is primarily based on an adaptive color model generation for hand segmentation followed by smoothing algorithm. We present a thinning algorithm followed by the construction of convex envelope to detect possible points for the fingertips. We demonstrate that the correct points for the fingertips can be selected heuristically through interpoints distance calculation. Finally, we show the effectiveness of the proposed method by experimenting with images of different background complexity and have achieved very promising results.

Product Type : Academic Conferences

Author : Ahmad Yahya Dawod, Faculty of Information Technology, Multimedia

PDF 4p

Languages : English

Partition Fusion Approach of Image Fusion Using Modified Pulse Coupled Neural Network for Dif

Company Description : The paper introduces an approach for fusion of differently focused images for improved human perception and machine recognition. The fusion algorithm uses a partition fusion approach where source images are Partioned into blocks and based on their clarity these blocks are selected for composite image. The quality of composite image thus depends upon the clarity measure and decision making system to select a block from given multi focus images. In the proposed algorithm the clarity of the image blocks is decided by EOL. The selection of blocks is performed by the Modified PCNN based on the clarity of the image blocks. To improve the performance of PCNN some modification are used in feeding and linking filed in PCNN. The performance of the proposed fusion algorithm has been tested for RMSE, Mutual Information and standard deviation. Considerable improvements are observed in the results as compared to existing method of image fusion.

Product Type : Academic Conferences

Author : Dheeraj Agrawal, Electronics and Communication Engineering Depar

PDF 6p

Languages : English

Identification of Patients with Obstructive Sleep Apnea Using Wavelets Packets and Artificial...

Company Description : 􀊚A new identification method for identification of patients with obstructive sleep apnea (OSA) from normal controls is investigated in this paper using estimated spectral analysis of RRI data with wavelets packets and artificial neural networks. Two sets of data are used in this paper. The training data is obtained from Sultan Qaboos University hospital while the test data is obtained from MIT databases. The training data set consists of 15 OSA and 15 normal subjects. The test data set is divided into two test sets each consists of 20 OSA and 10 normal subjects. The spectral analysis of RRI data obtained using 8 different sub-bands from wavelets packets is used as a classification feature. A simple artificial neural network of the type feed-forward back-propagation is used for the classification task. Different types of wavelets are used to test the consistency of the approach. The accuracy of classification approaches 92.7% using a large size of data simulated with power spectral density values of the main 8 sub-bands within the mean value plus/minus the standard deviation of the power spectral density values of the original test data sets.

Product Type : Academic Conferences

Author : Abdulnasir Hossen, Member IEEE, Department of Electrical and Com

PDF 6p

Languages : English

Fast Fractal Encoding through FFT using Modified Crosscorrelation based Similarity Measure

Company Description : The image compression using fractal transform is a promising method which is potentially capable of achieving very high compression ratios. The major drawback of fractal inage compression is large encoding time, though the decoding time is negligible. In this paper, a new similarity measure based on normalized cross-correlation of mean subtracted range and domain blocks is proposed.The fast fractal encoding algorithm based on the proposed similarity measure is well suited for FFT based frequency domain operations to speed up the encoding process. The implemented algorithm employs exhaustive search of similar domain blocks for each range block unlike other limited domain search methods. The algorithm works largely in frequency domain and operates on entire domain image instead of overlapping domain blocks.The contrast and brightness parameters of fractal transformation are easily calculated during the course of computation of similarity index matrix. Though the proposed method shows little dB drop in Peak signal to noise ratio(PSNR) values, the encoding time is reduced considerably with average speedup factor of 30 as compared to the full search method.

Product Type : Academic Conferences

Author : S.B Dhok,R.B.Deshmukh,A.G. Keskar, Visvesvaraya National Institu

PDF 5p

Languages : English

Intelligent Resource Selection for Sensor-Task Assignment: A Knowledge Based Approach

Company Description : Sensing resources play a crucial role in the success of critical tasks such as surveillance. Therefore, it is important to assigning appropriate sensing resources to tasks such that the selected resources fully cater the needs of the tasks. However, selecting the right resources to tasks is a computationally hard problem to solve. Most of the existing approaches address the efficiency aspect of the resource selection by considering the physical aspects of the sensor network (e.g., range, power, etc.) but have ignored important domain related properties such as capabilities of assets, environmental conditions, policies and so on which makes the selection effective. In this paper we present a knowledge rich mechanism to intelligently select ressources for tasks such that the selected resources sufficiently cover the needs of the tasks. Ontologies are used to capture the crucial domain knowledge and semantic matchmaking is used to perform sensor-task matching. A combination of ontological and first-order-logic reasoning is considered for the solution architecture.

Product Type : Academic Conferences

Author : Geeth de Mel, Wamberto Vasconcelos, Timothy J. Norman, Departmen

PDF 7p

Languages : English

Multi-click dependent model to estimate document relevance in web search

Company Description : Web search click logs, reflecting whether users are satisfied with the search results, are the most extensive and invaluable information resources of user preference. A central problem in click log analysis is to estimate the userperceived relevance of each query-URL pair. Many click models have been proposed to solve this problem, but they all have a common problem: the examine-next probability only depends on the current result or the preceding last clicked result. Intuitively, whether a user will continue to see the next result is supposed to be determined by whether one is satisfied with the information got from one’s historical clicked results, not only the current result. Therefore, we propose the multi-click dependent model (MCDM) that takes all the preceding clicked results into consideration. In the new model, the examine-next probability is decided by the click variables of each clicked result. We evaluate the proposed model on a real-world data set consisting of about 3.02 million query sessions obtained from a Chinese commercial search engine Sougou to test the performance of MCDM. The experiment results show that MCDM outperforms the existing click models in metrics such as log-likelihood, click perplexity, last click prediction error, especially on less-frequent queries and bottom positions of query sessions.

Product Type : Academic Conferences

Author : Xinyi Shu , Yujiu Yang†, Wenhuang Liu, Graduate School at Shenzh

PDF 7p

Languages : English

On Fuzziness in Hybrid Network Device Discovery

Company Description : The importance of network discovery cannot be denied especially for network monitoring and network management purposes. Here, we present a fuzzy approach to perform hybrid network device discovery. The main goal is to solve the increased additional traffic load issues in active discovery technique and the low accuracy issues in passive discovery technique. Then to further otptimize the hybridization between the passive and active approach, a fuzzy values are set to alternate the use of active and passive technique efficiently. Our preliminary results show that the proposed work are able to discover devices within the network-in-test with less additional traffic but with high accuracy.

Product Type : Academic Conferences

Author : Mustafa Abdat , Pradeep Isawasan , Muhammad Fermi Pasha†, Ahmed

PDF 5p

Languages : English

Neuro-Symbolic Integration using pseudo inverse rule

Company Description : Pseudo inverse learning rule is examined for its ability to accelerate the performance of doing logic programming in Hopfield neural network. This learning rule has a higher capacity than the Hebb Rule and Direct learning rule. This learning rule also suffers significantly less capacity loss as the network gets larger and more complex. Comparisons are made between these three rules to see which rule is better or outperformed other rules in the aspects of computation time, memory and complexity.

Product Type : Academic Conferences

Author : Saratha Sathasivam & 1Muraly Velavan; School of Mathematical Sci

PDF 5p

Languages : English

Identifying People’s Intention from Natural Language Texts

Company Description : This research is aimed at analyzing intention from ordinary language texts and producing a computer model of human agents, their intentions and actions. It will focus on words such as ‘assault’ which is the semantic key to recognize the people’s intentions. The input texts to analyze intention are from online legal databases. To ensure that the model of intention analysis is language independent, it will be developed and tested in languages from two different linguistic groups, English and Thai. An approach is proposed that merges machine learning with ontological analysis.

Product Type : Academic Conferences

Author : Sirichai Triamlumlerd & Dr.Jeremy Ellman

PDF 1p

Languages : English

Generation of Anaphoric Noun Phrases and their coreferential properties

Company Description : This paper investigates the generative processes involved in the formation of noun phrases (NPs) from an existing clause in a discourse. A speaker or writer is able to generate a range of NPs from a stated clause to co-refer to the different semantic units of the clause by combining the various primitive components. These generative processes can be grouped into either predicate-deletion or normalization. We use relations defined by Levi [1] as a basis to define constraints used in the formation of NPs by predicate-deletion and empirically determine the conformance level of these constraints with natural discourses. For the normalization process, we argue that the inflicted morpheme’s anaphoric properties can be characterized by their suffixes. We also show that, these anaphoric properties are independent of the choice between subjective and objective modifiers. The anaphoric properties of suffix-based morphemes is also empirically investigated for conformance levels, and the results are encouraging.

Product Type : Academic Conferences

Author : Parma Nand, School of Computing and Mathematical Sciences, Auckl

PDF 6p

Languages : English

Factors that Influence the Memory of Entertainment Experiences

Company Description : When generating an entertainment experience, such as that of a movie, a novel, or a computer game, not only the experience itself, but also the memory of the experience plays an important role. Pleasurable memories can generate additional pleasure during recall, be instrumental to generate secondary social pleasure when communicating the experience, and be important for word-of-mouth advertising of the entertainment product. In order to optimize entertainment experiences, especially for an online dynamic generation of experiences within computer games, we need to build formal models of these processes. This paper is a summary of a research study we conducted with 50 university students in order to identify factors that influence the memory of entertainment experiences. Our findings show that when watching a short film, a range of factors, including the level of attention of the person, the predictability of the storyline, and the valence of the content, all have substantial impact on the amount of information the test subject can recollect at a later stage. Psychophysiological measurements like skin conductance are also found to be indicators of the memory performance. In terms of the application of the results, some of the factors suggest the use of profiling techniques to best match the targtet audience's background, e.g., when used for automated story generation within a game.

Product Type : Academic Conferences

Author : Shenglan Kang and Alexander Nareyek

PDF 8p

Languages : English

Using Hash Tables to Expedite Knowledge Reasoning in the General Game Playing Agent

Company Description : General game playing research focuses on designing automated agents that accept declarative logic description of arbitrary games at run time and are able to play efficiently without human intervention. The game information including the game states, the rules of the game, and the player's role in the game are all represented in logic relations. The general game playing agent uses knowledge representation and reasoning algorithms to analyze and play the game. We use hash table to significantly improve the reasoning performance. We provide experimental data on seven different games: small games like the single player game Maze, the strategy game Mini-Chess, the two player game Tic-Tac-Toe, the middle size board games Connect Four and Chess Endgame, the large size game like Othello, and finally the three-player eCommerce game Farmer. In all of these scenarios, our agent has proven to significantly outperform the standard published Java player.

Product Type : Academic Conferences

Author : Xinxin Sheng, David Thuente, Computer Science Department, North

PDF 6p

Languages : English

Use of Probabilistic State Diagrams for Robot Navigation in an Indoor Environment

Company Description : This paper describes the analysis of syntax and semantics of state diagrams to support probabilistic behavior of robots. As a result we created probabilistic state diagrams that can be used for automatic code generation. The integration of the advanced robot vision algorithms with probabilistic state diagrams is also addressed. It is shown how to use techniques such as state abstraction and transition abstraction to create, verify and combine large probabilistic state diagrams. The paper also describes the implementation model for probabilistic state diagrams.

Product Type : Academic Conferences

Author : Bogdan Czejdo and Sambit Bhattacharya, Department of Mathematics

PDF 6p

Languages : English

Pleasure Differences as a Result of Seeing an Action versus Own Acting

Company Description : Actively playing a computer game will generate more pleasure than a corresponding passive experience, such as watching a movie or reading a book. This is what common knowledge tells us. It is however not fully consistent with our research on the human reward system, and we undertook the study described in this paper to have a closer look at the differences. In this study, test participants watched videos of actors playing games as well as actively played the games themselves. The study examined how the pleasure difference of the participant's own acting versus pasively warching the actions of an actor changes with the participant's level of empathy, the sympathy toward the actor, and the level of emotion displayed by the actors. Among other results, the findings indicate that a higher pleasure resulting from own actions diminishes with a higher empathy rating, with equal pleasure at about the maximum rating of 80.

Product Type : Academic Conferences

Author : Xinying Cheah and Alexander Nareyek

PDF 6p

Languages : English

PPNMF: Improving Weighted Nonnegative Matrix Factorization with Prior Information

Company Description : Collaborative Filtering (CF) is one of the most popular methods for recommendation problem. The key idea is to predict further the interests of a user (ratings) based on the available rating information from many users. Recently, matrix factorization(MF) based approaches, one branch of collaborative filtering, have proven successful for the rating prediction issues. As is well known, most of the MF models follow one of the following frames: (1) to fit a linear factor model over all observed ratings respect to the Frobenius norm with a positive regularization item, (2) to fit a linear factor model over all observed ratings respect to the Frobenius norm with a nonnegative constraint added. Differing from the exiting MF models, which model only on the observed ratings, the proposed variant of MF, referred to as Probabilistic Prior Nonnegative Matrix Factorization (PPNMF) in this paper, utilizes the prior information of the missing elements via treating each missing as a random variable; the probability distribution of each element is calculated with some specific scheme. Compared with the traditional Matrix Factorization for Collaborative Filtering, our empirical studies show that the proposed algorithm makes more accurate predictions of user ratings and is more robust with respect to the initial setting. With this method, we analyze the mechanism of its learning process, which shows that the algorithm will first go through a damping vibration process to make an adjustment, and then converges.

Product Type : Academic Conferences

Author : Shuli Han 1, Yujiu Yang 2, Wenhuang Liu 3, 1,2,3Graduate School

PDF 6p

Languages : English

Solving Multi-Objective Problems under Its Objective Bounds by Genetic Algorithms

Company Description : In this paper, a new approach is presented to solve the particular problems in which some objective functions are controlled to be within its objective bounds. The proposed algorithm called GABound algorithm is based on genetic algorithms (GAs) for searching the multi-objective solutions. The algorithm employs the preemptive optimization technique by considering one objective at a time based on the priority of the objective functions. The simulation results indicate that the proposed algorithm is satisfactory with customization of the number of eras and immigration rate.

Product Type : Academic Conferences

Author : Anon Sukstrienwong, Department of Information Technology, Bangko

PDF 6p

Languages : English

PID-GPSO Load Frequency Controller for Interconnected Power System

Company Description : This paper is proposing a Load Frequency (LF) controller design for interconnected power system based on Proportional Integral Differential (PID) controller. The PID gains have been design using Guided Particle Swarm Optimization (GPSO) technique. The controller is designed to improve the dynamic response of system frequency and the tie line power flow under a sudden load changes. The LFC model for a two area interconnected power system is implemented in SIMULINK/MATLAB and the simulation results illustrate the effectiveness of the proposed LF controller compared to integral controller.

Product Type : Academic Conferences

Author : Amer S. Al-Hinai, Sultan Qaboos University, P.O. Box 33 Al-Khodh

PDF 7p

Languages : English

Organizer : Global Science & Technology Forum

GSTF provides a global intellectual platform for top notch academics and industry professionals to actively interact and share their groundbreaking research achievements. GSTF is dedicated to promoting research and development and offers an inter-disciplinary intellectual platform for leading scientists, researchers, academics and industry professionals across Asia Pacific to actively consult, network and collaborate with their counterparts across the globe.