Re: [tesseract-ocr] How to get the correct text orientation with tesseract

[Zdenko Podobny]

script detection was always problematic and tesseract try to identify only a few...

Regarding rotation you can get better results by using the language file:

>tesseract unnamed.jpg - --psm 0 -l jpn
Warning, detects only orientation with -l jpn
Estimating resolution as 262
Warning. Invalid resolution 0 dpi. Using 70 instead.
Page number: 0
Orientation in degrees: 90
Rotate: 270
Orientation confidence: 6.44
Script: Han
Script confidence: 1.43

Zdenko

pi 10. 3. 2023 o 18:21 nguyen ngoc hai <nguyenng...@gmail.com> napísal(a):

I have the following image:

 

I used the following code to get the text orientation, it works for most of my samples except the above image.

```python
    def get_orientation_confidence(cv2_img_data):
        image = cv2pil(cv2_img_data)
        osd_result = {}

        with tesserocr.PyTessBaseAPI(lang='osd') as api:
            api.SetImage(image)
            api.SetSourceResolution(300)

            osd_result = api.DetectOrientationScript()

        return osd_result

    # preprocess image before detecting orientation
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    gray_white_border = self.make_border_white(gray)
    self.show_image("gray_white_border", gray_white_border)

    # Threshold the image to convert it to black and white
    threshold = cv2.threshold(gray_white_border, 0, 255, cv2.THRESH_OTSU)[1]
    self.show_image("threshold otsu", threshold)

    osd_ret = get_orientation_confidence(pre_roi_im)
    print(osd_ret['orient_deg'])
```
```cmd
{'orient_deg': 180, 'orient_conf': 0.06795501708984375, 'script_name': 'Arabic', 'script_conf': 0.0}
```
Here, the results I got were not correct, and also wrong language detection. 

I hope to get {'orient_deg': 90, 'script_name': 'Japanese', ...}
I supposed the results belonged to tesseract's output results.

Is that possible to get the correct orientation degree here?
Assuming that I already know the language, are there any methods (such as applying extra image preprocessing, etc.) that can provide better accuracy here?

Thank you very much for your time.
I hope to hear any suggestions.

--
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-oc...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/e447f23e-a0e1-4a91-b6e1-0eca8511f7acn%40googlegroups.com.

[nguyen ngoc hai]

Dear Zdenko,

Thank you very much for your suggestion.

May I ask which version of tesseract are you using?
I ran the same command with tesseract v5.0.0, but I got a different result. 

```
>tesseract -v
tesseract v5.0.0-alpha.20210811

...

Warning, detects only orientation with -l jpn

Page number: 0
Orientation in degrees: 270
Rotate: 90
Orientation confidence: 46.00
Script: Latin
Script confidence: 2.00

```
Should I upgrade to the newest version of tesseract or try some extra preprocessing methods before detecting text orientation?
Thank you for your time. 
Best regards
Hai

You received this message because you are subscribed to a topic in the Google Groups "tesseract-ocr" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/tesseract-ocr/CPTtW5bPqYc/unsubscribe.
To unsubscribe from this group and all its topics, send an email to tesseract-oc...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CAJbzG8xoY%2BTVbQLuSXXN3u-5LEAPpZ4nq7CJHdFRXLQJta2yBQ%40mail.gmail.com.

--

Nguyen Ngoc Hai

Phone:  +81 1488 4168  (JP).

skype ID: nguyenngochaibkhn.

[Zdenko Podobny]

the latest code (5.3.0) (on windows)

Zdenko

so 11. 3. 2023 o 2:16 nguyen ngoc hai <nguyenng...@gmail.com> napísal(a):

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CA%2BWjAfqTWpZ5rbkAUFVY2-cKhKBFq3CY33bAaCyVLtv3tsGWXw%40mail.gmail.com.

[nguyen ngoc hai]

Thank you very much for your help.

I will give it a try. 

Best regards

Hai 

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CAJbzG8wZDotnyN8NpGpbDPPrpWG7vDJj_sX6XrOZAUsfa888qw%40mail.gmail.com.

[Zdenko Podobny]

one more thing: I used a language file from https://github.com/tesseract-ocr/tessdata e.g. with legacy engine data.

Zdenko

so 11. 3. 2023 o 13:18 nguyen ngoc hai <nguyenng...@gmail.com> napísal(a):

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CA%2BWjAfoP4JY4%2BLEfAKvA2qrua86jh5jf6KWJoaMoBiL2hvp_Jg%40mail.gmail.com.

[nguyen ngoc hai]

Dear Zdenko and everyone, 

Thank you for your help last time. 

Apologize for getting back a bit late, I could get the same results by using the same language which you suggested. 
However, the language model gave me less accurate OCR results than the language model in tessdata_best.
It is troublesome, but would it be possible for tesseract to specify a different model (from the same language)?
For example:
Use the Legacy model for OSD, and use the tessdata_best model for extracting text.

Please also forgive me that due to the data privacy matter, I will have to delete the uploaded image from the post later. 

Thank you for your time. 
Best regards
Hai