How to get digital and the confidence?

[朱裕清]

This is my target image:


Actually my question is similar to [this post](https://stackoverflow.com/questions/4944830/how-to-make-tesseract-to-recognize-only-numbers-when-they-are-mixed-with-letter). But I don't know why the following answer will lead to another direction. I mean, I just hope to get those digits with high degree of confidence. Such as I can do this with another language


Then I can just keep those degree of confidence with a threshold `0.9`. But now I hope to use *Tesseract* to do this.

First, I train a *number.traineddata* just for recognizing number. You can get it [here](https://1drv.ms/u/s!Aumb0ijJibxOi1KVXFjwDzOVRQrm).

   

tesseract.exe target.jpg stdout -l number --oem 0 -psm 6


Note I will get all digits which include high confidence and low confidence. Can we recognize the number and get the degree of its confidence? I cannot find any information to implement it. If *Tesseract* cannot do it. Any other method based on **C++** can implement my target? Could anyone can give me some information for it?

[ShreeDevi Kumar]

Your image is 96 dpi. Increase the dpi to 300 and try.

Preprocess the image to remove the boxes around letters, if possible.

See https://github.com/tesseract-ocr/tesseract/wiki/ImproveQuality

ShreeDevi
____________________________________________________________
भजन - कीर्तन - आरती @ http://bhajans.ramparivar.com

--
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+unsubscribe@googlegroups.com.
To post to this group, send email to tesser...@googlegroups.com.
Visit this group at https://groups.google.com/group/tesseract-ocr.
To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/273d9f86-39ce-42fe-8934-781f2103e4fa%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[朱裕清]

I think that is another question. I want to filter recognized characters by confidence in this question but don't know how to get the confidence with Tesseract.

在 2017年10月20日星期五 UTC+8下午4:13:35，shree写道：

Your image is 96 dpi. Increase the dpi to 300 and try.

Preprocess the image to remove the boxes around letters, if possible.

See https://github.com/tesseract-ocr/tesseract/wiki/ImproveQuality

ShreeDevi
____________________________________________________________
भजन - कीर्तन - आरती @ http://bhajans.ramparivar.com

On Fri, Oct 20, 2017 at 1:24 PM, 朱裕清 <zyd1...@gmail.com> wrote:

This is my target image:


Actually my question is similar to [this post](https://stackoverflow.com/questions/4944830/how-to-make-tesseract-to-recognize-only-numbers-when-they-are-mixed-with-letter). But I don't know why the following answer will lead to another direction. I mean, I just hope to get those digits with high degree of confidence. Such as I can do this with another language


Then I can just keep those degree of confidence with a threshold `0.9`. But now I hope to use *Tesseract* to do this.

First, I train a *number.traineddata* just for recognizing number. You can get it [here](https://1drv.ms/u/s!Aumb0ijJibxOi1KVXFjwDzOVRQrm).

   

tesseract.exe target.jpg stdout -l number --oem 0 -psm 6


Note I will get all digits which include high confidence and low confidence. Can we recognize the number and get the degree of its confidence? I cannot find any information to implement it. If *Tesseract* cannot do it. Any other method based on **C++** can implement my target? Could anyone can give me some information for it?

--
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.

[Art Rhyno.]

Maybe look at the API [1]. The output of the attached program shows there’s a lot of detail that can be gleaned at this level, including the confidence of the selected character and that of the other candidates. Compiling against tesseract on Ubuntu, at least, is fairly straightforward. I don’t know about windows or os/x.

 

art

---

1. https://github.com/tesseract-ocr/tesseract/wiki/APIExample

To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/790f9169-e724-49b2-b24a-320a10fea6f4%40googlegroups.com.

[朱裕清]

Sorry for late response. And thank you very much. You solution help me a lot.

在 2017年10月21日星期六 UTC+8上午11:57:48，Art Rhyno写道：

[朱裕清]

Hello, This is my target_image.tif  

 

I have changed your code a little to be this

#include <tesseract/baseapi.h>
#include <leptonica/allheaders.h>
//#include<opencv.hpp>
//using namespace cv;
using namespace std;


int main()
{
 tesseract::TessBaseAPI *api = new tesseract::TessBaseAPI();


 if (api->Init(".\\tessdata", "eng")) {
 fprintf(stderr, "Could not initialize tesseract.\n");
 exit(1);
 }
 api->SetPageSegMode(tesseract::PSM_SINGLE_BLOCK);


 //Mat  mat_image = imread("target_image.tif", 0);
 Pix *image = pixRead("target_image.tif");
 //cvtColor(mat_image, mat_image, CV_GRAY2BGR);
 api->SetImage(image);
 api->SetSourceResolution(300);
 //Important
 api->Recognize(NULL);


 tesseract::ResultIterator* ri = api->GetIterator();
 tesseract::PageIteratorLevel level = tesseract::RIL_SYMBOL;


 //Mat image_rect_bin(mat_image.size(), CV_8UC1, Scalar(0));
 int line = 0;
 if (ri != 0) {
 do {
 const char* symbol = ri->GetUTF8Text(level);
 if (ri->IsAtBeginningOf(tesseract::RIL_TEXTLINE))
 line++;


 if (symbol != 0) {
 int x1, y1, x2, y2;
 ri->BoundingBox(level, &x1, &y1, &x2, &y2);
 tesseract::ChoiceIterator ci(*ri);
 do {
 const char* choice = ci.GetUTF8Text();
 //rectangle(image_rect_bin, Point(x1, y1), Point(x2, y2), Scalar(255), -1);
 } while (ci.Next());
 }
 delete[] symbol;
 } while (ri->Next(level));
 }
 api->End();


 return 0;
}

Then I think I can get all rectangle(The rectangle is important for me.). But if I use opencv to show those rectangle I have got. it will be this.

As you see, I will get so many dis-separated rectangle from the target_image.tif. Could you help me again? Please...

I think there is a threshold to control this behavior. But I cannot find anything from the documentation..