Unable to detect text in few images when using multi-language hints

[t6nand]

I am using tesseract-OCR to extract text from images which can contain text in many popular Indian languages (it can also be bi-lingual or multi-lingual in the same image).
While trying multi-language hints in my command, I often encounter error
contains_unichar_id(unichar_id):Error:Assert failed:in file ../ccutil/unicharset.h, line 513

On manually checking that image and providing that language hint works correctly.

Command resulting in error:
tesseract --tessdata-dir /usr/local/share/tessdata/ tamil.jpg -l hin+eng+tam+guj+pan+kan+mal+ben+tel+mar stdout

Correctly working command:
tesseract --tessdata-dir /usr/local/share/tessdata/ tamil.jpg -l tam stdout

Affected Sample Image:

Environment

• Tesseract Version:
  tesseract 4.00.00alpha
  leptonica-1.74.1
  libjpeg 8d : libpng 1.6.32 : libtiff 4.0.8 : zlib 1.2.8
  Found AVX2
  Found AVX
  Found SSE
• Commit Number: ebbfc3a
• Platform: Darwin - Kernel Version 16.7.0 (MacOS)

Current Behavior:

Using multi-language hints in command results in error:
contains_unichar_id(unichar_id):Error:Assert failed:in file ../ccutil/unicharset.h, line 513

Expected Behavior:

It should extract text matching the language profiles provided in command without error.

Suggested Fix:

Since directly using text's language profile works correctly, it may have to do something in providing multi-language hints. So, in providing multiple languages this behavior is unexpected and may need correction.

[lloiodice]

Found same problem running one of the UNLV dataset images
tesseract 9444_011.2B.tif 9444_011.2B.tif

As work-around, Modified code ../ccutil/unicharset.h, line 513
From
ASSERT_HOST(contains_unichar_id(unichar_id));
To
if (!contains_unichar_id(unichar_id)) return false;

After that I was able to OCR the image and get valid text for it

[IamDixit]

I had a similar issue.
Got it solved by taking a new training set from https://github.com/tesseract-ocr/tesseract/wiki/Data-Files#updated-data-files-for-version-400-september-15-2017

[amitdo]

@t6nand,

Please retest it with 4.0.0-rc3

[Shreeshrii]

Retest with

tesseract 4.1.0-rc1-255-g332a1
 leptonica-1.76.0
  libgif 5.1.4 : libjpeg 8d (libjpeg-turbo 1.4.2) : libpng 1.2.54 : libtiff 4.0.6 : zlib 1.2.8 : libwebp 0.4.4 : libopenjp2 2.3.0

Command as originally given does not work - stdout needs to be given after image name

$ tesseract indic.jpg -l hin+eng+tam+guj+pan+kan+mal+ben+tel+mar stdout
read_params_file: Can't open hin+eng+tam+guj+pan+kan+mal+ben+tel+mar
read_params_file: Can't open stdout
Tesseract Open Source OCR Engine v4.1.0-rc1-255-g332a1 with Leptonica
Warning: Invalid resolution 0 dpi. Using 70 instead.
Estimating resolution as 350
Detected 38 diacritics

No assert when multiple language hints given

 tesseract indic.jpg stdout -l hin+eng+tam+guj+pan+kan+mal+ben+tel+mar
Warning: Invalid resolution 0 dpi. Using 70 instead.
Estimating resolution as 334
Detected 44 diacritics
0000390 ಪ್ರೇ) நீங்கள்‌ நினைவில்‌
ம வேண்டிய ஒரு உண்மை
_ இந்த நிமிடம்‌கூட நிரந்தரமில்லை

3 ந

ഛീ

Recognition is different when only one language is given

tesseract indic.jpg stdout -l tam
Warning: Invalid resolution 0 dpi. Using 70 instead.
Estimating resolution as 350
Detected 38 diacritics
றன்‌ இமான்‌ நீங்கள்‌ நினைவில்‌
ம வேண்டிய ஒரு உண்மை

இந்த நிமிடம்‌?கூட நிரந்தரமில்லை
5) நத

க நட
த.

The suggested approach could be to use --psm 0 to identify the script and then use appropriate language data.

 tesseract indic.jpg stdout --psm 0
Warning: Invalid resolution 0 dpi. Using 70 instead.
Estimating resolution as 350
Page number: 0
Orientation in degrees: 0
Rotate: 0
Orientation confidence: 2.51
Script: Tamil
Script confidence: 3.89

[Shreeshrii]

Also, using --psm 6 gives better recognition.
Image has 3 lines of text but very colorful and busy background.

 tesseract indic.jpg stdout -l tam --psm 6
Warning: Invalid resolution 0 dpi. Using 70 instead.
க்‌ க்‌
இன்பத்திலும்‌" துன்பத்திலும்‌ மட்டி)
வைத்துக்கொள்ளவேண்டிய ஒரு உண்மை
இந்த நிமிடம்‌ கூட நிரந்தரமில்லை
டு வு த
ஜு
ந்‌ கு]
த்‌ வ்‌
8 ்‌
ubuntu@tesseract-ocr:~/TEST$ tesseract indic.jpg stdout -l tam --psm 6 --dpi 300
க்‌ க்‌
இன்பத்திலும்‌" துன்பத்திலும்‌ மட்டி)
வைத்துக்கொள்ளவேண்டிய ஒரு உண்மை
இந்த நிமிடம்‌ கூட நிரந்தரமில்லை
டு வு த
ஜு
ந்‌ கு]
த்‌ வ்‌
8 ்‌

[stweil]

@t6nand, is this issue solved for you?

[t6nand]

@stweil Hi, I can't ascertain if it's solved as this was a long time ago and back then, I worked on some workaround to get the best-approximated results. For now, I will have to set up the latest tesseract-ocr and try out the suggestions as discussed on this thread. I will post my findings in some days. Thanks.

[t6nand]

@stweil, @Shreeshrii .

The original issue seems to have been resolved as there's no contains_unichar_id error. However, the intelligibility of OCRed text is quite different from using multi-language hints when compared to using only a single hint as also discussed by @Shreeshrii in his suggestion. Also, I receive a segmentation fault issue when using the command tesseract indic.jpg stdout --psm 0.

[Shreeshrii]

I receive a segmentation fault issue when using the command tesseract indic.jpg stdout --psm 0.

I tested just now with current master and do not get the segfault. Please post full details of your setup.

tesseract indic.jpg stdout --psm 0

Warning: Invalid resolution 0 dpi. Using 70 instead.
Estimating resolution as 350
Warning. Invalid resolution 0 dpi. Using 70 instead.
Page number: 0
Orientation in degrees: 0
Rotate: 0
Orientation confidence: 2.51
Script: Tamil
Script confidence: 3.89

tesseract -v

tesseract 5.0.0-alpha-322-g74ac
 leptonica-1.78.0
  libgif 5.1.4 : libjpeg 8d (libjpeg-turbo 1.4.2) : libpng 1.2.54 : libtiff 4.0.9 : zlib 1.2.11 : libwebp 0.4.4 : libopenjp2 2.3.0