A Harmony Search Algorithm for Recognition-Based Segmentation of Online Arabic Text

In this paper a Harmony Search algorithm (HS) is used for online Arabic text recognition. The algorithm is divided into two phases: text segmentation using dominant point detection and character recognition using HS. The segmentation algorithm uses dominant point detection to mark minimal number of points which could form the text skeleton. Then, the generated text skeleton is expressed as a directional model with 6 directions. This directional model minimizes the directions opposite to the writing direction. As a result, the new text directional expression will exploit all the possible segmentation points. Finally, HS is used to match the best database character to the target character generated from the segmentation process by minimizing the total score obtained from the overall text matching. The system is tested using a database of 4500 words forming 21234 characters in different positions or forms (isolated, start, middle and end). The data set is divided into a set of 3000 words for training and 1500 words for testing. The algorithm scored a 93.4% successful word recognition rate with an execution time of 4.3 sec.