I found a single quite simple function from Leptonica that its 
output(compiled for ios) differ to the regular version.
The function is pixRotateBinaryNice .. 
Digging in the function I think the cause for discrepancy is one of the 
following:

1) different floating point behavior.
2) different implementation of sin cos function
3) possible undefined behaviour of shift left operation on a signed integer 
with possible overflow
(line 165 rotateamlow.c    xp = xcen + (xpm >> 4);)  
xpm is a signed integer running from -w/2 to w/2 where w is the image width.


I don't have direct access to an ios development platform(Mac) so if 
someone can verify this it could help.

Alternatively someone could try adding some -fsanitize   flags to the build 
script in Robert Mathews github
By the way Robert excellent work and I appreciate it very much that you 
share it.




   

On Sunday, March 2, 2014 1:57:51 PM UTC+2, Tsur Herman wrote:
>
> I am experiencing the same issues.. and found difference in Leptonica also 
> from regular version to ios version.
> I don't have alue yet what is causing this .. maybe compiler optimization? 
> floating point definition?
>
> did anybody find something  .. help! :-)
>
> On Tuesday, January 7, 2014 11:48:47 AM UTC+2, Aleksander Grzyb wrote:
>>
>> Thank you very much for your project, I downloaded it and tested and 
>> unfortunately results for me are still different...
>>
>> Could you take look of a sample I am using (image is after preprocessing):
>>
>>
>> <https://lh6.googleusercontent.com/-FiiBgjL_wec/UsvLPUJBXII/AAAAAAAAAMU/67Frq1lGkGk/s1600/1.JPG>
>>
>> I am using polish language to detect words in the above image, I 
>> downloaded the language data from: 
>> https://code.google.com/p/tesseract-ocr/downloads/detail?name=tesseract-ocr-3.02.pol.tar.gz&can=2&q=
>>
>> I application I set language to "pol", "tessedit_pageseg_mode" to "2" and 
>> I don't resize the image, because it is already small (514x104) and the 
>> output from console command (tesseract 1.JPG output -l pol) gives me result 
>> "PARAGON FISKALNY" which is perfect and iOS appliaction gives me "PHRHGON 
>> FISKHLNY" which is really strange because every letter "A" changes to "H". 
>> I think my next step is to create file "unicharambigs" and use 
>> combine_tessdata, but it is very strange that console version gives me 
>> different result than iOS version, I tried every parameter and tried to 
>> resize photo, but with no results. Do you have any ideas what can cause 
>> problem?
>>
>>
>>
>>
>>
>>
>>
>> W dniu niedziela, 5 stycznia 2014 03:50:39 UTC+1 użytkownik Robert 
>> Mathews napisał:
>>>
>>> read my thread more carefully. i did recompile against tesseract 3.02
>>>
>>> Typos courtesy of my iPhone
>>>
>>> On Jan 4, 2014, at 6:15 PM, Benjamin Sølberg <[email protected]> 
>>> wrote:
>>>
>>> Hi Robert
>>>
>>> You probably already know this but your project uses an old 
>>> version/snapshot of tesseract.
>>> Just a heads up as I was hoping that that you were using the latest code 
>>> :-)
>>> There have been at least one fix regarding the osx version.
>>>
>>> Benjamin
>>>
>>> Den fredag den 3. januar 2014 21.20.27 UTC+1 skrev Robert Mathews:
>>>>
>>>> I recompiled against the latest tesseract and leptonica-1.69
>>>>
>>>> You can see the project I used to compile here: 
>>>> https://github.com/robmathews/compile-tesseract
>>>>
>>>> Then, I updated the sample ios app to 
>>>> - use tesseract 3.02 + leptonica-1.69
>>>> - allow choosing a photo from the photo library
>>>>
>>>> and checked into this fork: 
>>>> https://github.com/robmathews/OCR-iOS-Example
>>>>
>>>> And that's all I know.
>>>>
>>>>
>>>>  -- 
>>> -- 
>>> You received this message because you are subscribed to the Google
>>> Groups "tesseract-ocr" group.
>>> To post to this group, send email to [email protected]
>>> To unsubscribe from this group, send email to
>>> [email protected]
>>> For more options, visit this group at
>>> http://groups.google.com/group/tesseract-ocr?hl=en
>>>  
>>> --- 
>>> You received this message because you are subscribed to the Google 
>>> Groups "tesseract-ocr" group.
>>> To unsubscribe from this group and stop receiving emails from it, send 
>>> an email to [email protected].
>>> For more options, visit https://groups.google.com/groups/opt_out.
>>>
>>>

-- 
-- 
You received this message because you are subscribed to the Google
Groups "tesseract-ocr" group.
To post to this group, send email to [email protected]
To unsubscribe from this group, send email to
[email protected]
For more options, visit this group at
http://groups.google.com/group/tesseract-ocr?hl=en

--- 
You received this message because you are subscribed to the Google Groups 
"tesseract-ocr" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/groups/opt_out.

Reply via email to