DEVELOPMENT OF AN ALGORITHM AND GUIDELINES FOR THAI WRITING VERIFICATION USING IMAGE PROCESSING FOR UPPER PRIMARY SCHOOL STUDENTS: A CASE STUDY OF BANSANPARAI SCHOOL

Abstract

This research aims to 1) Apply Zipf’s Law in the analysis of Thai vocabulary, 2) Apply language patterns in word classification to examine Thai writing, 3) Develop a system for checking Thai sentence writing, and 4) Propose guidelines for using image processing-based Thai writing verification through quantitative research.  The studied data consisted of Thai writing samples from 37 upper-elementary students (late primary level) at Ban San Parai School, Tak Province. Research instruments included a Thai vocabulary test, a writing test, an image-processing-based writing verification system, descriptive statistics, model performance testing, and expert validation. 

The research findings were as follows: 1) The application of Zipf’s Law in analyzing Thai vocabulary knowledge enabled the classification of words into 4 groups based on quartiles: most frequently written words, easy words, medium words, and difficult words. 2) The application of a Trigram language model combined with handwriting recognition and Convolutional Neural Network was able to classify words from handwritten images and correctly reorder them to match the original sentences with 100% accuracy, even when the input data had uncertain ordering. 3) The developed algorithm for checking Thai sentence writing achieved an Accuracy, Precision, Recall, and F1-score = 1.0 on both training and test sets. There were no errors in the Confusion Matrix for both Sentence 1 and Sentence 2. 4) The proposed guidelines for using image-processing-based Thai writing verification were deemed highly appropriate, with an average suitability score of 4.40 (on a 5-point scale), corresponding to a “high” level of appropriateness.