Convert image for best recognition?

[Viton-zizu]

Excuse me.
How i can transform my image for better recognition?
I have image with Russians chars, but when i recognize picture, i have realy bad result

Bitmap pictureInfoArea = src.ToBitmap();
TesseractEngine engine = new TesseractEngine("tessdata/", "rus", EngineMode.Default);
var page = engine.Process(pictureInfoArea, PageSegMode.Auto);
string s = page.GetText();

probably, i can teaching tesseract for best result?
sorry for my worse english
This black text in tesseract
[spoiler]
РОССИЙСКАЯ
Паспорт выдан „
ФЕДЕРАЦИЯ
отдвлвнивм угас госсии_„щ
В Р-НЕ ГОР.
А 1ёДо2.2оо9
дата выдачи, ‚
.. ‚ Кодуподратздежения '
27овоо4Т$
ёо во
[/spoiler]
This black text true
[spoiler]
ОТДЕЛЕНИЕМ УФМС РОССИИ
ПО ХАБАРОВСКОМУ КРАЮ
В КОРОВСКОМ Р-НЕ ГОР ХАБАРОВСКА
11.02.2002
[/spoiler]

[clockwiseq]

Are you sure you have the language files in the tessdata folder for rus? That was my problem and after adding the language files, it pulled the proper text. Verify that the language files are there and let’s go from there.

From: Viton-zizu [mailto:notifications@github.com]
Sent: Monday, August 25, 2014 9:51 PM
To: charlesw/tesseract
Subject: [tesseract] Convert image for best recognition? (#115)

Excuse me.
How i can transform my image for better recognition?
I have image with Russians chars, but when i recognize picture, i have realy bad result

Bitmap pictureInfoArea = src.ToBitmap();
TesseractEngine engine = new TesseractEngine("tessdata/", "rus", EngineMode.Default);

var page = engine.Process(pictureInfoArea, PageSegMode.Auto);
string s = page.GetText();

probably, i can teaching tesseract for best result?
sorry for my worse english

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/115.

[charlesw]

No I don't, I've only included English for testing purposes. It's up to you
to download and install the languages you need as stated in the projects
readme.
On 26 Aug 2014 13:33, "clockwiseq" notifications@github.com wrote:

Are you sure you have the language files in the tessdata folder for rus?
That was my problem and after adding the language files, it pulled the
proper text. Verify that the language files are there and let’s go from
there.

From: Viton-zizu [mailto:notifications@github.com]
Sent: Monday, August 25, 2014 9:51 PM
To: charlesw/tesseract
Subject: [tesseract] Convert image for best recognition? (#115)

Excuse me.
How i can transform my image for better recognition?
I have image with Russians chars, but when i recognize picture, i have
realy bad result

Bitmap pictureInfoArea = src.ToBitmap();
TesseractEngine engine = new TesseractEngine("tessdata/", "rus",
EngineMode.Default);

var page = engine.Process(pictureInfoArea, PageSegMode.Auto);
string s = page.GetText();

probably, i can teaching tesseract for best result?
sorry for my worse english

—
Reply to this email directly or view it on GitHub<
https://github.com/charlesw/tesseract/issues/115>.

—
Reply to this email directly or view it on GitHub
#115 (comment).

[clockwiseq]

Did he ever respond and say that he did not have the language files? Also, I'll never respond to the emails again because I didn't know it included everything in the thread. Oh well :-)

[charlesw]

Sorry clockwiseq I got a bit muddled up there, must have been a bit tired.
Please ignore my response, what you say is a good point reading it a second
time :)
On 27 Aug 2014 03:05, "clockwiseq" notifications@github.com wrote:

Did he ever respond and say that he did not have the language files? Also,
I'll never respond to the emails again because I didn't know it included
everything in the thread. Oh well :-)

—
Reply to this email directly or view it on GitHub
#115 (comment).

[Viton-zizu]

clockwiseq
yes, off course, rus data files! 3.02 and some chars recognize ok, but not all. How i can do better?

[charlesw]

Are you doing any image pre-processing? From the looks of it you'll need to do a bit to get reasonable results. Checkout out http://stackoverflow.com/questions/23361583/how-to-use-opencvtesseract-for-accurate-text-recognition-in-android. I'll see if I can come up with a pre-processing sample in the meantime as this would probably be useful to others as well. However I probably won't have time till the weekend.

[charlesw]

Excellent! I finally had a chance to play around with some pre processing
techniques over the weekend and found the following to give me the best
results:

1. Convert to grayscale
2. Subtract the medium blur (block size ~letter size) from the image
3. Binarize the image

That being said it still missed some letters and the process could be
further improved by removing straight lines.

A complementry approach might be to also do some post processing such as
using a dictionary or potential matching the results against a database if
possible.
On 29 Aug 2014 15:26, "Viton-zizu" notifications@github.com wrote:

Yes, i do! Gaussian blur-> convert to gray -> Threshold
For example:
[image: image]
https://cloud.githubusercontent.com/assets/8552019/4086691/f14b2d30-2f3c-11e4-8149-39b3041b7977.jpg

—
Reply to this email directly or view it on GitHub
#115 (comment).

[Viton-zizu]

Great! Thank you!

[maxmoore14]

@charlesw Can the image pre-processing you describe be done server-side with .NET? Are there built in functions in System.Drawing perhaps? Or do you know of an open source library that could help? I'm trying to make the process as seamless as possible for my end users who will be uploading files.

[etishor]

@maxmoore14 we use the AForge ( http://www.aforgenet.com/framework/documentation.html ) library, available on .nuget, for pre-processing images. Depending on images you might want to play with the contrast, gray-scale and binarization of the source images.

Another thing to note, if you need to scale, and run on multiple cores ( 8+ ), using System.Drawing.Bitmap is not recommended as it is using GDI+ which holds a process level lock for every operation causing a lot of contention.

[maxmoore14]

@etishor Thanks for the quick response. I'm trying to build a system that will allow users to upload various types of documents and parse the text. For my users, they are likely going to be JPGs from iPhones & Androids. Often the contents will be handwritten notes, but not always. I am very inexperienced with image processing in general, so if you have some guidelines or suggestions on some specific functions I should call, filters I should apply, etc., that would be really helpful.

[etishor]

@maxmoore14 AForge is very easy to use, take a look at their documentation & samples.