The Clustering Technique for Thai Handwritten Recognition
Ithipan Methasate, Sutat Sae-Tang · 2004
This paper describes an algorithm for clustering freestyle Thai handwritten character models. The algorithm groups the characters that have a similar structure. Firstly, the algorithm begins with the vertical stroke detection. The vertical stroke is an important Thai character structure. Secondly, the character area is divided into 7/spl times/10 blocks by using the stroke information. Then, the pixel distribution feature is calculated from each block. The features are trained using backpropagation neural network. Finally, the confusion matrix is used to analyze the result in a clustering process. The characters are divided into 21 groups and the accuracy of the clustered model is 97.60 percent.