Ilya Mezhirov

No, I mean that often letters form a single connected component, and minidjvu is entirely helpless against that. Somebody who owns a proprietary djvu encoder (I forgot who) wrote that his encoder is integrated with an OCR system in non-trivial ways, in particular, the OCR helps to split single connected components that consist of several letters, and also it prevents misidentifying letter images that are recognized differently by the OCR. There's very little minidjvu can do against that (unless we...

In this terms the suggestion to compare with "union of the first 500 patterns" is seems to be an approximation of a dot cluster with only its central point (centroid) What I'm talking about is a balltree (omohundro.files.wordpress.com/2009/03/omohundro89_five_balltree_construction_algorithms.pdf) in the space of binary images, also called hit-and-miss tree, and it is exact, not an approximation. By "union" I don't mean any kind of averaging, just OR (and AND for the complement). The approximation...

С вашего позволения No problem. Как я понимаю, наработки по составлению локального jb2 словаря были в итоге интегрированы в djvulibre? Djvulibre makes its own jb2 dict. The minidjvu part of djvulibre gets an array of shapes and returns the substitutions. I think that speeding up multipage compression requires not just parallelization, but abandoning the quadratic search algorithm that minidjvu currently uses. The shape matcher considers letters A and B equivalent if (1) the skeleton of A is a subset...

Hi Alexander! I'll answer in Russian since you want to post on linux.org.ru. Историю разработки расскажу, почему бы и нет. Надо было писать курсовую по математике, а нормальная математическая тема не подвернулась. В таких случаях обычная практика - написать какую-нибудь полезную программу по усмотрению научного руководителя, и он предложил улучшить djvu. Первоначально предполагалось, что другие студенты тоже что-нибудь поделают, в частности поэтому я стал переписывать части djvulibre на C - djvulibre...

And authors of this code seems to be unreachable now. Really?! Thank you for doing this. In recent years I didn't have any time for this at all, there were a bunch of CVEs against minidjvu that I'd probably need to fix first. (aggression command line option) that I don't know how to optimize. Neither did I. The basic equivalence is simply this: two letters are equivalent if the skeleton of one is a subset of the dilation of the other and vice versa. However, that's way too aggressive so the rest...