An effective approach to offline Arabic handwriting recognition

• A spatial representations for Arabic characters are proposed.
• A new approach to the skeletonization of handwriting images documents is introduced.
• Experiments were performed on the IFN/ENIT databases.
• The approach is successful even when using handwritten upper case English characters.

Segmentation is the most challenging part of Arabic handwriting recognition due to the unique characteristics of Arabic writing that allow the same shape to denote different characters. An Arabic handwriting recognition system cannot be successful without using an appropriate segmentation method. In this paper, a very effective and efficient off-line Arabic handwriting recognition approach is proposed. The proposed approach has three stages. Firstly, all characters are simplified to single-pixel-thin images that preserve the fundamental writing characteristics. Secondly, the image pixels are normalized into horizontal and vertical lines only. Therefore, the different writing styles can be unified and the shapes of characters are standardized. Finally, these orthogonal lines are coded as unique vectors; each vector represents one letter of a word. To evaluate the proposed techniques, we have tested our approach on two different datasets. Our experimental results show that the proposed approach has superior performance over the state-of-the-art approaches.

Figure optionsDownload as PowerPoint slide