Segmentation of Hanacaraka Characters Using Double Projection Profile and Hough Transform

Abstract

In doing segmentation of Hanacaraka character, Javanese ancient character, one of Indonesian’s ethnic ancient character in Java island, the difficulties that occur is the inconsistency of the space between lines, the size of the character and the thickness. Inconsistencies between row spacing and letter size are caused by the letters of the pair, the last vowel and consonant letters in one phoneme. While the thickness is inconsistent due to the writing style of the Hanacaraka itself.

Image Preprocessing needs to be done to get input without skew. To improve skewed text documents, we used Hough transforms to predict the edges of the text area. After that, to segment the line and then continue with segmentation of each character, horizontal projection profile is used and then proceed with vertical.

The result of this segmentation method is good for printed documents. Segmentation process of handwriting documents has difficulty because each row in the document is uneven and very tight between the rows. Those matters cause them overlap. When the line segmented wrongly, the entire character on the line will be not segmented as well. This problem can be eliminate using connectivity test. Before this, it need to segment the line with the overlap area. The character part of below or above the main character can be eliminate because it is not connected to the main character.