There are two problems here. In Arabic-based writing systems, dots play a significant role, much more so than in Latin-based scripts. Therefore, these dots should be classified correctly.

The second problem is that the connected components are not always reliable. There are unduly divided components which are part of a single component. We can’t label them as they are, and uniting them into a uniform component requires manual intervention—something we try to avoid.

In the coming days, I’ll try to exemplify these problems and how we treat them.

Keywords:

• ridge
• printed text
• non-text segmentation
• gaussian-filter bank
• reading order

Q1: How is line skew determined?

There is a $\theta$ parameter in the Gaussian kernel which is used to produce ridges. This may be used in detecting the skew, but since it’s constant for an entire page, a varying line skew will probably decrease its performance.

Q2: How are non-text portions detected?

The paper does not include a description but cites “S. S. Bukhari, F. Shafait, and T. M. Breuel, ‘Improved document image segmentation algorithm using multiresolution morphology,’ in Proc. SPIE Document Recognition and Retrieval XVIII, San Jose, CA, USA, Jan. 2011” as a source for an improved technique.

Q3: Which heuristics are used in reading order determination?

Breuel is reported to have an algorithm in “T. M. Breuel, ‘High performance document layout analysis,’ in Symposium on Document Image Understanding Technology, Greenbelt, MD, USA, April 2003.” The paper says the authors modified the algorithm for right-to-left scripts. No further details are provided.

Q4: How large is the dataset, and what does it contain?

25 Arabic documents and 20 Urdu documents are used.

Q5: Are there any techniques applicable to divans?

There might be, if any of them were described in detail. We already have more sophisticated text line detection techniques. For the others, I’ll need to read the cited works.