Greetings,
I found a potential solution to rewrite each pixel to either white or black
based on a set threshold. After looking at OpenCV functions I found
"threshold" would do just that but Tesseract was still finding "ghost"
characters in the white areas of the image. So I had to find where the
string starts and grab an ROI from that point. Note that the
THRESH_BINARY_INV parameter to threshold will also convert dark colors to
white and light colors to black. From things I've read Tesseract likes
black characters on white backgrounds.
So the solution I came up with is the following using OpenCV and tesseract:
Mat img; // should already have the image
Mat cropped;
Mat grayed;
Mat inverted;
Mat cropNum;
// Crop the original image to the defined ROI
Rect roi(xStart,yStart,xMove,yMove);
cropped = img(roi);
// Convert Image to Gray
cvtColor(cropped, grayed, COLOR_BGR2GRAY);
// Invert Image to black and white
threshold(grayed, inverted, 100, 255, THRESH_BINARY_INV);
// Use tesseract to OCR
tesseract::TessBaseAPI *ocr = new tesseract::TessBaseAPI();
ocr->Init(NULL, "eng", tesseract::OEM_LSTM_ONLY);
ocr->SetPageSegMode(tesseract::PSM_SINGLE_WORD);
ocr->SetImage( inverted .data, inverted .cols, inverted .rows, 1,
inverted .step);
popupNum = string(ocr->GetUTF8Text());
NOTE: Be careful with the 4th parameter in ocr->SetImage function.
This is the number of bits per pixel.
After converting to grayscale it's 1 and not 3. I forgot
about this and I was getting 3 strings back. Quite strange.
On Thursday, February 24, 2022 at 11:02:27 PM UTC-7 Ed Dow wrote:
> Greetings,
>
> I'm using tesseract 4.0.0 in a C/C++ application where I capture an image
> and then "scrape" text/data from it. I am having issues with tesseract
> recognizing the ROI with just several characters ( see attached).
>
> The attached image is: *014*
> Recognized as: */~—6h014 5*
>
> If I get rid of extra space around the number it gets better but the
> problem is sometimes the string of characters is outside the ROI so I have
> to increase the size to get all of them.
>
> I've tried using OpenCV to grayscale, blur and resize which has seemed to
> help a little. I've also tried all the PSM modes.
>
> The other thing that is puzzling is that from the command line it works
> great. Maybe this is due to the image being saved as a jpg first before
> the OCR is done. Inside the application it's raw data.
>
> Any thoughts?
> Ed
>
>
> Tesseract Version:
>
> tesseract 4.0.0-beta.1
> leptonica-1.75.3
> libgif 5.1.4 :
> libjpeg 8d (libjpeg-turbo 1.5.2) :
> libpng 1.6.34 :
> libtiff 4.0.9 :
> zlib 1.2.11 :
> libwebp 0.6.1 :
> libopenjp2 2.3.0
>
--
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/bae3383d-84ee-402c-aa2f-af4fe7273a4fn%40googlegroups.com.
