Abdul Rahim Ahmad Azizah Suliman Nur Shakirah Md

Abdul Rahim Ahmad, Azizah Suliman, Nur Shakirah Md Salleh abdrahim@uniten. edu. my, azizah@uniten. edu. my, shakirah@uniten. edu. my DESCRIPTION The invention is a handwriting recognition system engine. The system was initially developed using French words and English words database collected by IRESTE, University of Nantes in France. It is adapted for Malay words recognition by retraining using Malay cheque word database collected at UNITEN, which we call UCW (UNITEN Cheque Words) database. In handwriting recognition system, a Hybrid of Neural Network (NN) and Hidden Markov Model (HMM) in word modeling is known to give good recognition than discrete Hidden Markov Model (HMM). However, in using NN, the use of Empirical Risk minimization (ERM) results in poor generalization. Aiming to improve generalization and recognition, we use Support Vector Machine (SVM) in place of NN. SVM uses structural risk minimization (SRM) which allows simultaneous optimization of representational and discriminative capability. We evaluated SVM in character recognition using isolated character databases of IRESTE and the hybrid SVM/HMM using UCW word databases. Overall Concept of the System The Recognition Process BENEFIT The handwriting recognition engine can be used in an application that require recognition of Malay cheque words. The engine can handle on-line handwriting signal in which the recognition is done using the coordinates of the movement of the pen. Thus, the application that uses this engine need to use the online pen such as Anotto to capture the pen movement coordinates. The signal captured can be processed immediately to be fed to an accounting spreadsheet or stored in the memory of the pen for future summary report. The Training and Recognition of the Character Recognizer The Overall Training process of the Word Recognizer COMMERCIALISATION NOVELTY The novelty of this invention is in using a different method than normal to perform the recognition. In most handwriting recognition engine, Neural Network (NN) is normally used for recognizing segmentation of word into individual characters and Hidden Markov Model (HMM) is used to combined the appropriate segments to form the complete word recognition. In our invention, we used Support Vector Machine (SVM) instead to replace NN. There is a potential for commercializing the product. The product is a result of an experiment on the usage of SVM to replace NN in an already commercial product My. Script Builder by Vision Objects. Even though the engine gives high accuracy of recognition, due to somewhat bigger model size, more work need to be done to find out the solutions for it before it can really be used in the commercial version of the product. TEAM MEMBERS ABDUL RAHIM AHMAD AZIZAH SULIMAN NURSHAKIRAH MD. SALLEH Department of Systems and Networking, College of Information Technology, Universityi Tenaga Nasional, Km 7, Jalan Kajang-Puchong, 43009 Kajang, Selangor. Malaysia. ABDUL RAHIM AHMAD AZIZAH SULIMAN NURSHAKIRAH MD. SALLEH