Data and Software engineering Office Research Profile

Information and computer science department research profile l.jpg
1 / 56
0
0
1413 days ago, 485 views
PowerPoint PPT Presentation
Data and Computer Science Department Research Profile. Dr. Wasfi Al-Khatib Information and Computer Science Department King Fahd University of Petroleum and Minerals. Data and Computer Science Faculty. 25 Professorial Rank employees 1 Full Professor 5 Associate Professors

Presentation Transcript

Slide 1

´╗┐Data and Computer Science Department Research Profile Dr. Wasfi Al-Khatib Information and Computer Science Department King Fahd University of Petroleum & Minerals

Slide 2

Information and Computer Science Faculty 25 Professorial Rank employees 1 Full Professor 5 Associate Professors 19 Assistant Professors 2 PhD. Holders 1 Instructor 1 Lecturer

Slide 3

ICS Research Areas Computer Vision, Image Audio and Video Processing and Arabization. Computerized reasoning: Theorem Proving, Software and equipment Verification, machine learning, design acknowledgment, Uncertainty and information Reasoning Computer Networks: Network outline, Performance and Optimization, Mobile and Distributed Computing Systems, High-Speed Networks, Sensor Networks, Active Networks. Working Systems: OS for Mobile gadgets, Distributed Systems, Multi-Agent Systems, Multimedia Systems, Computer Security. Programming Engineering: Object-situated Software Engineering, Software Design, Software Measurements Computer Science Education and eLearning. PC Algorithms: Parallel Computing, Computational Geometry, Randomized Algorithms, Grid Computing, Web-mining, information mining. Database Systems: Database Design, Query Optimization, XML Databases, Multimedia Databases

Slide 4

ICS Research Projects: Computer Vision, Image, Audio, and Video Processing Towards the Further Study of Designing with NURBS & ANURBS: The CAD/CAM/CAE Tools, KFUPM/SABIC, 2002-2004. Programmed Text Recognition: A Need in Arabization, KFUPM, 2001-2005 Automatic Font Generation: A stage ahead in Arabization, KFUPM, 2000-2002 Automatic Classification of music and discourse in digitized sound.

Slide 5

Recognition of License Plates Objectives Identification of auto number plates in complex foundation. Plate extraction from poor pictures with low difference, glare influenced power profiles and movement obscure. Character Segmentation from plates at high tilt or picture skew. Acknowledgment under a lead base relevant to Saudi Arabian number plate authorizing benchmarks. Foundation of a standard number plate database that doesn't exist for Saudi Arabia (Arabic plates) right now. Improvement of novel and advanced procedures in the space of Image Processing, Computer Vision and Machine Learning.

Slide 6

Proposed and Implemented Approach Mainly includes three stages: Extraction, Segmentation and Recognition Achievements Contrast Adjustment utilizing Histogram Stretching Local element extraction in view of common picture edge profiles and break lights. Plate extraction utilizing changed Fuzzy Vector/Euclidean edge identification based methods. Character division utilizing a bi-group Fuzzy C-implies calculations Recognition of fragmented character bitmaps utilizing PCA.

Slide 7

Development potential outcomes with Intel A useful License Plate Recognition (LPR) System requires superb Image Grabbing Devices for operations that are, something else, extremely tedious under programming test systems. A LPR framework by and by is a part of an Intelligent Transport System. Bolsters various PC groups for continuous information connect. Complex picture preparing undertakings are performed in parallel utilizing different PCs. Various such operations that are worked over equipment progressively frameworks are Frame Averaging Image Differencing Color level transformations Edge and Intensity Adjustment operations. Division and Labeling Significant work has been done in the acknowledgment of US, EU, Japanese, Korean and Chinese License plates. A down to earth License Plate Recognition (LPR) System is required NOT only for the nearby need of the Kingdom additionally at a worldwide level which can incorporate different other Arab nations in the neighbor. The venture will be valuable for different applications and can be utilized to uphold speed confines on freeways/streets, screen movement streams at activity signals, record stopping measurements at parts, auto burglary observing, Border Crossing, and so on

Slide 8

Arabization Projects Arabic Text-to-Speech (ATTS) Two sorts of discourse units were utilized autonomously: The main comprises of 375 diphones of Arabic sounds, and alternate has 178 allophones which cover Arabic and English sounds. The venture created broad Arabic semantic instruments including: Arabic articulation standards, and tables of sporadically professed Arabic words, and allophone/diphone determination rules. A parametric model was additionally worked to blend the discourse and to give the client control over the pitch rate, stress, and discourse beat.

Slide 9

Arabization Projects Automatic Generation of the Arabic Diacritical Marks We detailed the issue of producing Arabic diacritized content from unvoweled content utilizing Hidden Markov Models (HMM) approach. The word grouping of unvoweled Arabic content is viewed as a perception succession from a HMM, where the shrouded states are the conceivable diacritized articulations of the words. The ideal arrangement of diacritized words (or states) are then gotten effectively utilizing a Viterbi like Algorithm. The principal period of this venture has as of now accomplished 94.5% letter precision.

Slide 10

Arabization Projects Arabic Speech Recognition System The venture goes for building adequate know how and a best in class inquire about base for the improvement of the cutting edge discourse acknowledgment procedures for the Arabic dialect. This venture utilizes Carnegie Mellon University Sphinx-II, Sphinx-III and Sphinx-IV ASR motors and devices as a base for building a cutting edge huge vocabulary speaker-free consistent Arabic ASR frameworks. The venture includes building a substantial Arabic discourse corpus, an Arabic phonetic lexicon, preparing Arabic triphone parametric models, and improvement of broad instruments for displaying Arabic common dialect. The venture is executed mutually with the Center of Speech and Phonological Science at King Abdulaziz City of Science and Technology. Target application: Automatic TV/Radio news translation.

Slide 11

Arabization Projects Neural Network based Speech acknowledgment. The proposed extend goes for researching different structures for ANN/HMM models for phoneme acknowledgment or cutting edge Arabic Speech acknowledgment. Carnegie Mellon Sphinx-4 will be utilized as our testing stage.

Slide 12

Automatic Classification of Speech and Music lessening/expulsion from documentaries Speech Segments Extraction Automatic discourse acknowledgment Indexing and recovery Speaker acknowledgment Improving sound coding/pressure Feature Selection Classifier Features Speech Music Feature Extraction Process Classification Process

Slide 13

Automatic Classification of Speech and Music: Methodology Newly Proposed Features RMS of Lowpass Signal Mean of Discrete Wavelet Transform (DWT) Variance of Discrete Wavelet Transform (DWT) Range of Zero Crossings Variance of Mel Frequency Cepstral Coefficients (MFCC) Previously Used Features Spectral Flux Percentage of Low Energy Frames Linear Predictive Coefficients (LPC) Contribution of removed elements considered utilizing Fuzzy C-Means Clustering Classification Frameworks Neural Networks Multilayer Perceptron (MLP) Radial Basis Function (RBF) Statistical Models Hidden Markov Model

Slide 14

Automatic Classification of Speech and Music: Prototype System

Slide 15

ICS Research Projects: Artificial Intelligence Learning Prolog programs: hypothesis and applications in information mining. Basic Assessment of Key Analytical Methods for Sanding Prediction. 2005-2006. Create Fuzzy Logic Models to Generate Permeability Traces in Non-Cored Wells. 2005-2006. Improvement of Artificial Intelligence System for Prediction and Quality Control of PVT Properties. 2005-2006. Multi-Agent Based Ubiquitous Approach for Personalized Information Systems.

Slide 16

6 Decision Region 1 Region 2 5 4 3 2 x 2 1 Decision 0 Boundary - 1 2 3 4 5 6 7 8 9 10 x1 FUNCTIONAL NETWORKS AS A NEW FRAMEWORK FOR PATTERN RECOGNITION Computer Science Approaches: 1. Bolster Vector Machines 2. Probabilistic Neural Network (NN) 3 . Outspread Basis Functions Network 4. Multilayer Perceptron NN Statistical Approaches: 5. Discriminant Analyses 6. Strategic Regression 7. K-Nearest Neighbor Functional systems are a speculation of neural systems . They are cape capable of captur ing & speaking to complex info/yield connections.

Slide 17

Functional Networks Classifier Learning Algorithm We accept that the likelihood can be composed as: where are obscure, yet unhindered capacities to be gained from the information, and p (.) must fulfill the two likelihood conditions, and is questions. For instance, p (.) can be a Probit or Sigmoidal or CDF or Mulinomial strategic capacities. In utilitarian systems, we learn capacities (not parameters) by surmised them by directly autonomous family: The parameters can be scholarly utilizing enhancement strategies. The reaction is: We utilize Constrained Least Squares, or Iterative Least Squares , or Maximum Likelihood

Slide 18

Simulation and Real-World Applications of Functional Networks: A Comparative Study The genuine Databases under study are taken from: Machine learning vault database at UC Irvine : ftp://ftp.ics.uci.edu/bar/machine-learning-databases Thalassemias Data: p=4, c=3

Slide 19

Functional Networks: Internal and External Validation Using p=4 and c=3

Slide 20

Develop Fuzzy Logic Models to Generate Permeability Traces in Non-Cored Wells Carbonate rocks represent an extraordinary test for mapping rock properties, particularly porosity and penetrability, because of their variable and complex pore structure Fuzzy Logic affirms that the supply comprises of a few litho-sorts, each having trademark appropriations for porousness and electrical log values. Fluffy Logic endeavors to reveal the relationship between these circulations. O bjective: Develop ability in fluffy rationale porousness displaying that utilizations routine open-gap lo

SPONSORS