Using projection and loop for segmentation of touching Thai typewritten

This paper proposes a segmentation technique for touching Thai typewritten characters. Thai characters vary in size and position when they are in a sentence. A Thai word is composed of consonants, vowels and tones. Touching characters can occur both in horizontal and vertical directions. The proposed technique uses structural characteristics to detect suitable segmentation points in both directions. The segmentation process consists of four steps. First the height and then the position of characters are used to identify character zones. Next, size and both horizontal and vertical projections are used to classify the types of touching. Lastly, touching characters are segmented using directions and positions identified by the previous steps. The edge of touching characters is used to identify the edge of two isolated characters. The proposed segmentation technique is tested with both electronic typewriters and manual portable typewriters. Segmentation accuracy of 95.4% has been obtained for two hundred sentences of typewritten thesis documents.