Offline arabic handwritten text recognition: A Survey
ACM Computing Surveys (CSUR)
| Hi-index | 0.00 |
Slant correction is an important part of the normalization task in OCR applications. Due to some special specifications of Farsi and Arabic manuscripts, conventional deslanting methods proposed for other languages do not work properly. In this paper, a fast method is first introduced to estimate the overall tilt of a handwritten word based on directional filters. After overall deslanting, a novel non-uniform slant estimation algorithm computes the remaining slant of each near-vertical stroke of the word, separately. Each candidate stroke is traced and its slant is calculated. A non-uniform slant correction algorithm is also proposed to reduce the remaining slants of each candidate stroke keeping the distortions of other strokes of the word at a minimum level. Thanks to the special characteristics of Farsi/Arabic scripts, slants are estimated in a specific strip of the written words. A comparison between our approach and three other prevalent methods is drawn. Experiments show that the proposed overall slant estimation method not only represents the least estimation error, but is also the fastest algorithm. The best results are achieved using the proposed overall and non-uniform deslanting methods. It is concluded that successful results can be achieved by considering the special specifications of these two languages.