An evolutionary harmony search algorithm with dominant point detection for recognition-based segmentation of online Arabic text recognition

Ain Shams Engineering Journal

Umi Kalthum Ngah¹  Moayad Yousif Potrus²  Bestoun S. Ahmed² 

[1] School of Electrical and Electronic Engineering, Universiti Sains Malaysia, Engineering Campus, 14300 Nibong Tebal, Seberang Perai Selatan, Pulau Pinang, Malaysia;Software Engineering Department, Engineering College, Salahaddin University-Hawler (SUH), 44002 Erbil, Kurdistan, Iraq;

关键词: Character segmentation;    Dominant point detection;    Handwritten Arabic text recognition;    Harmony Search;    Genetic Algorithms;    Artificial intelligent;   

DOI  :  10.1016/j.asej.2014.05.003

来源: DOAJ

【 摘 要 】

This paper highlights a novel strategy for online Arabic text recognition using a hybrid Genetic Algorithm (GA) and Harmony Search algorithm (HS). The strategy is divided into two phases: text segmentation using dominant point detection, and recognition-based segmentation using GA and HS. At first, the pre-segmentation algorithm uses a modified dominant point detection algorithm to mark a minimal number of points which defines the text skeleton. The generated text skeleton from this process is expressed as directional vector, using 6-directional model, to minimize the effect of character body on segmentation process. Then, GA and HS algorithms are used as recognition-based segmentation phase for text and character recognition respectively. For the segmentation based recognition, binary GA is used to explore different combinations of segmentation points which gives the best score, while HS is integrated inside the GA segmentation to explore the best character score produced from matching the character with different characters stored in the database. In order to initially calibrate and test the system, a locally collected text dataset was used that contains 4500 Arabic words. The algorithm scored a 93.4% successful word recognition rate. Finally, the system was tested on the benchmark ADAB dataset 2 consist of 7851 Arabic words and it scored a successful recognition rate in the range of 94–96%.

【 授权许可】

Unknown