CONSTRUCTING METHOD OF HIDDEN MARKOV MODEL FOR ON-LINE KOREAN ALPHABET RECOGNITION

摘要

PURPOSE: A HMM(Hidden Markov Model) for on-line Korean alphabet recognition is provided to recognize Korean alphabet written in cursive and original grapheme. The HMM is constructed based on information of a stroke of a character. CONSTITUTION: A stroke of a character inputted on touch screen of HPD(Handheld Personal Computer) is converted into a x and y axis coordinates string. The string is divided into primitive stroke string representing the stroke and having a length and a direction. A template of conventional expanded basic strokes having a code and a primitive stroke string is stored in the HPC. A stroke recognition is completed by comparing the stroke inputted with the strokes in template and finding out a code of most similar stroke in the template. A location information between each input strokes is detected by defining directional relationship between an end point of the former stroke and a start point of the latter stroke as a relative location. An observation symbol string for the HMM is generated from combination of the code of the stroke and the directional relationship between the stroke and the other stroke. The HMM for Korean alphabet recognition written in cursive is constructed by modifying transition probability between each state in a conventional HMM for Korean alphabet recognition written in square form.