Fakhry Ikhsan Firdaus, Achmad Khumaini, Fitri Utaminingrum
The Arabic letters are very different from the latin. Arabic letters can be interconnected with other letters, in addition, the Arabic letters also have many variety of shapes based on its position. The important point to obtain high accuracy for recognizing latin Arabic letter is the segmentation process. This is a challenge when we will segment each letter. Especially if the letter is interconnected. Many researchers have proposed methods for segmenting Arabic writing in the previous research. In this paper, we propose Modified Connected Component Labeling (MCCL) to perform Arabic letter segmentation. Performance of Connected Component Labeling (CCL) in the previous research to separate each interconnected object still needs an improving. For under certain conditions there are still letters that are connected with other letters so that the letters are still not segmented correctly. To solve that problem, we apply MCCL with multiple rules based on moment location, object location, and height of the object to segment each letter, especially for letters that are connected with others. We evaluate our proposed method using quantitative analysis compared to CCL method. Our method successfully segmented each Arabic letter. Referring to the testing results, the performance of MCCL is better than CCL. © 2017 IEEE.
Faculty of Computer Science, Brawijaya University, Malang, Indonesia
