Thank you for the links and knowledge!
It definitely makes a good read, and a fine introduction to missing pieces.
Stay safe and healthy,
Ciobanita Paul Adrian.
~ SATCOM Sr. Systems Engineer / DevOps engineer / Test Engineer ~
Skype: adrian_iss_consult
On Wed, 31 Aug 2022 at 18:52, Zdenko Podobny <zde...@gmail.com> wrote:
> There is nothing like 100% OCR accuracy. Simply from a bad image you can
> not get good results (maybe google vision is close ;-), but it is a
> different story).
> Our best experiences are collected at docs (
> https://tesseract-ocr.github.io/tessdoc/ImproveQuality.html).
>
> For different images/problems you need different solutions. E.g. in the
> case of historical documents you will need to focus on thresholding, in the
> case of natural scenes on text detection, in the case of invoices, document
> layout processing...
>
> There is much research on this: some older papers are available on
> academia.edu E.g.
>
> https://www.academia.edu/4790395/PhotoOCR_Reading_Text_in_Uncontrolled_Conditions
> https://www.academia.edu/2793675/End_to_end_scene_text_recognition
> https://www.academia.edu/6030087/Tex_Binarization_In_Color_Documents
>
> https://www.academia.edu/39052965/OCR_to_read_embossed_text_from_Credit_Debit_card
>
> https://www.academia.edu/1171645/A_variational_approach_to_degraded_document_enhancement
>
> https://www.academia.edu/19957320/Multi_spectral_document_image_binarization_using_image_fusion_and_background_subtraction_techniques
>
> https://www.academia.edu/1171639/Low_quality_document_image_modeling_and_enhancement
>
>
> Zdenko
>
>
> st 31. 8. 2022 o 17:10 Adrian Paul Ciobanita <adrian.cioban...@gmail.com>
> napísal(a):
>
>> Can you recomend tutorials, or books avout how to do image pre-processing
>> effectively and efficiently?
>>
>> Do we need to do different types of image pre-processing for each image?
>> If we have 100+ images, how do we ensure that the pre-processing is helping
>> the prediction accuracy 100%?
>>
>> On Wed, Aug 31, 2022, 18:04 Zdenko Podobny <zde...@gmail.com> wrote:
>>
>>> Trained data from tesseract 5 are compatible with 4, so definitely I
>>> would suggest using the latest tesseract version for training - there was a
>>> lot of bug fixing and speed improvements.
>>>
>>> IMO tesseract training has never been easy. I always suggest focusing on
>>> image preprocessing rather than training.
>>> Following an easy looking training tutorial could also mean you are on
>>> the wrong path (=> you waste your time and it increases your frustration).
>>> For example:Tesseract 4 has 2 engines LSTM and legacy engine. In this
>>> particular video, for which engine is that training? For hints see[1]. Did
>>> you plan to train that engine or other one?
>>>
>>> [1]
>>> https://tesseract-ocr.github.io/tessdoc/tess5/TrainingTesseract-5.html#overview-of-training-process
>>>
>>> Zdenko
>>>
>>>
>>> st 31. 8. 2022 o 15:40 John Alway <jal...@gmail.com> napísal(a):
>>>
>>>> "First of all: if you follow any tutorial on internet - report the
>>>> problem to the author of the tutorial."
>>>> Next: use official documentation for training. I see there are a bunch
>>>> of folks just "generating content" - to gain an audience. Without insight
>>>> and therefore also without support, using old/outdated information..."
>>>>
>>>> People are trying to find a nice, easy tutorial to help them get
>>>> through the forest. I think that's the bottom line. Thanks for the
>>>> link.
>>>>
>>>> 'Tesseract 4 was released 29 Oct 2018. Almost 4 year ago! The recent
>>>> tesseract version is 5.2 and training process was also improved:
>>>> https://github.com/tesseract-ocr/tesstrain";
>>>>
>>>> I understand this, but I'm using C# .Net, and I don't think version 5
>>>> is available in C#. Unless I'm mistaken? There are costly packages, such
>>>> as IronOcr which uses tesseract 5, but there is no way I can take that
>>>> route.
>>>>
>>>> Regards,
>>>> ...John
>>>>
>>>>
>>>>
>>>> [image: width=]
>>>> <http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail>
>>>> Virus-free.www.avg.com
>>>> <http://www.avg.com/email-signature?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail>
>>>> <#m_2367304440822020296_m_-3692493932810461313_m_4818300399805832802_m_6502694561301915279_DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2>
>>>>
>>>> On Wed, Aug 31, 2022 at 3:27 AM Zdenko Podobny <zde...@gmail.com>
>>>> wrote:
>>>>
>>>>> First of all: if you follow any tutorial on internet - report the
>>>>> problem to the author of the tutorial.
>>>>> Next: use official documentation for training. I see there are a bunch
>>>>> of folks just "generating content" - to gain an audience. Without insight
>>>>> and therefore also without support, using old/outdated information...
>>>>> Tesseract 4 was released 29 Oct 2018. Almost 4 year ago! The recent
>>>>> tesseract version is 5.2 and training process was also improved:
>>>>> https://github.com/tesseract-ocr/tesstrain
>>>>>
>>>>> Zdenko
>>>>>
>>>>>
>>>>> st 31. 8. 2022 o 0:18 John Alway <jal...@gmail.com> napísal(a):
>>>>>
>>>>>> Hello,
>>>>>>
>>>>>> I've been following a tutorial on youtube titled "Tesseract OCR -
>>>>>> Lesson 2: Training Tesseract for new font" here:
>>>>>> https://www.youtube.com/watch?v=1v8BPw0Dn0I&ab_channel=TheCode
>>>>>>
>>>>>> I'm using tesseract 4.0 on Window 10.
>>>>>>
>>>>>> I went through the steps he used, and everything seems to go smoothly
>>>>>> until I get to the actual training. When I run "mftraining" the
>>>>>> program
>>>>>> hangs. It seems to get stuck and doesn't indicate why are what it's
>>>>>> doing.
>>>>>>
>>>>>> I'm using a set of fonts in an image. I have the full alphabet upper
>>>>>> and lower case and the numbers 0 to 9 on the png image. I've attached
>>>>>> the
>>>>>> image. Unlike him, I'm using the English. I don't know the font, so I'm
>>>>>> just calling it tiktok to give it a name. My training file is called
>>>>>> *eng.tiktok.exp0.tr
>>>>>> <http://eng.tiktok.exp0.tr> *
>>>>>>
>>>>>> I used* jTessBoxEditor* to correct mistakes and set the box sizes
>>>>>> and positions precisely.
>>>>>>
>>>>>>
>>>>>> When I run this command:
>>>>>> *mftraining -F font_properties -U unicharset -O eng.unicharset
>>>>>> eng.tiktok.exp0.tr <http://eng.tiktok.exp0.tr>*
>>>>>>
>>>>>> The program just hangs. I've waited over twenty minutes.
>>>>>>
>>>>>> Should I wait longer? What could cause it to hang?
>>>>>>
>>>>>>
>>>>>>
>>>>>> Thanks!
>>>>>> ...John
>>>>>>
>>>>>> --
>>>>>> 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 tesseract-ocr+unsubscr...@googlegroups.com.
>>>>>> To view this discussion on the web visit
>>>>>> https://groups.google.com/d/msgid/tesseract-ocr/534c3f74-420b-4c96-83dd-609bcb002f81n%40googlegroups.com
>>>>>> <https://groups.google.com/d/msgid/tesseract-ocr/534c3f74-420b-4c96-83dd-609bcb002f81n%40googlegroups.com?utm_medium=email&utm_source=footer>
>>>>>> .
>>>>>>
>>>>> --
>>>>> 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 tesseract-ocr+unsubscr...@googlegroups.com.
>>>>> To view this discussion on the web visit
>>>>> https://groups.google.com/d/msgid/tesseract-ocr/CAJbzG8w5MX1XchQp6jfu2Vz06zWp82HxbDHrgp7%2BQ_Neh%2BDeug%40mail.gmail.com
>>>>> <https://groups.google.com/d/msgid/tesseract-ocr/CAJbzG8w5MX1XchQp6jfu2Vz06zWp82HxbDHrgp7%2BQ_Neh%2BDeug%40mail.gmail.com?utm_medium=email&utm_source=footer>
>>>>> .
>>>>>
>>>> --
>>>> 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 tesseract-ocr+unsubscr...@googlegroups.com.
>>>> To view this discussion on the web visit
>>>> https://groups.google.com/d/msgid/tesseract-ocr/CAN7TTkEYLexQwmN9y3qO6cwqGoW72HoaEj8XksAgTwb3qVSNPA%40mail.gmail.com
>>>> <https://groups.google.com/d/msgid/tesseract-ocr/CAN7TTkEYLexQwmN9y3qO6cwqGoW72HoaEj8XksAgTwb3qVSNPA%40mail.gmail.com?utm_medium=email&utm_source=footer>
>>>> .
>>>>
>>> --
>>> 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 tesseract-ocr+unsubscr...@googlegroups.com.
>>> To view this discussion on the web visit
>>> https://groups.google.com/d/msgid/tesseract-ocr/CAJbzG8wt1fjKy66uZ9-sEAqUkEwagLAZhFxcDV_8QLuPgk%3DnvQ%40mail.gmail.com
>>> <https://groups.google.com/d/msgid/tesseract-ocr/CAJbzG8wt1fjKy66uZ9-sEAqUkEwagLAZhFxcDV_8QLuPgk%3DnvQ%40mail.gmail.com?utm_medium=email&utm_source=footer>
>>> .
>>>
>> --
>> 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 tesseract-ocr+unsubscr...@googlegroups.com.
>> To view this discussion on the web visit
>> https://groups.google.com/d/msgid/tesseract-ocr/CADB4xcjLwGXFqNFnxwFVWcOXAtztmd3aLrs4G6AnzXawWYwf_A%40mail.gmail.com
>> <https://groups.google.com/d/msgid/tesseract-ocr/CADB4xcjLwGXFqNFnxwFVWcOXAtztmd3aLrs4G6AnzXawWYwf_A%40mail.gmail.com?utm_medium=email&utm_source=footer>
>> .
>>
> --
> 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 tesseract-ocr+unsubscr...@googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/tesseract-ocr/CAJbzG8woFXFYfo9zHQ0Vj%3D0JrqJPSWphq7C7WGF9YLUOeHwDFg%40mail.gmail.com
> <https://groups.google.com/d/msgid/tesseract-ocr/CAJbzG8woFXFYfo9zHQ0Vj%3D0JrqJPSWphq7C7WGF9YLUOeHwDFg%40mail.gmail.com?utm_medium=email&utm_source=footer>
> .
>
--
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 tesseract-ocr+unsubscr...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/tesseract-ocr/CADB4xciOsizZzZtpVeHyhgWuJu2u5MDJe5eeGBsXqKY7D8j-Sg%40mail.gmail.com.
