Search code examples
phpimage-processingimagemagickocrtesseract

Tesseract - Detecting small font size of image and convert to text

I have a screenshot of the bank cheque, I need all the text from this screenshot but tesseract is unable to read it properly. I also tried to pre-process the image but the output fails miserably.

I am using ImageMagick for pre-processing and Tesseract for recognizing text. The link to the image: https://i.sstatic.net/uNNhS.jpg

I am able to retrieve the account number, but not IFSC code and person name "SRINIVAS"

The steps I am following is as follows: 

magick -density 300 check1.jpg -depth 8 -strip -background white -alpha off check1.png

magick convert check1.png -resize 250% res_check1.png

convert -brightness-contrast 10x30 res_check1.png b_res_check1.png

convert b_res_check1.png -threshold 45% bin_res_check1.png

tesseract bin_res_check1.png o_res_check1

Note: I tried to resize the image upto 400% but it did not work. Google Vision API is able to read and convert every single text properly.

Solution

  • In ImageMagick, you can use -lat (local area threshold) to process the image to clean the background. I also have a bash shell script, text cleaner, at http://www.fmwconcepts.com/imagemagick/index.php.

    Input:

    enter image description here

    For ImageMagick 6, try

    convert input.jpg -negate -lat 25x25+10% -negate result.png

    Vary the -lat arguments to see if you can improve the results.

    For ImageMagick 7, replace convert with magick.

    enter image description here

    Does that help? Small fonts are going to be hard to process. Enlarging a raster image, does not usually help much. But you can try with sharpening filters or sharper resize filters, such as -filter catrom. See http://www.imagemagick.org/Usage/filter/

    You can also try doing a perspective transformation to rectify the original image before further processing. This might also help. See http://www.imagemagick.org/Usage/distorts/#perspective