I am very new to python and image processing, so i want to ask whether captcha like these below here are solvable by processing (e.g. with opencv) and then using google's tesseract (or another ocr/ml process) to read them or not? To be exact, i want to know how to make these images more readable by tesseract.

maybe you can first remove the long colored lines that is obscuring the characters and then turn the image to black/white high contrast image? separate each characters by using the large white space between them? that would surely make things easier for OCR to do its work.

If you start by mapping all the different chars you can find and assigning labels to them, you can use a CNN to solve the problem with pytorch, smth like the models used for imagenet.

Captcha generally poor implemented like it will be implemented like post request does not send the Captcha, or you can bypass it. Tried that first.

Next thing I recommend  what captcha code (I mean the service like breached also used bb2 captcha) is used or get them, use OCR( you can use the existing code made by someone or make your  own (I will prefer this If you can do it)) and train yourself a model.

(August 11, 2022, 10:43 PM)dthang Wrote: If you start by mapping all the different chars you can find and assigning labels to them, you can use a CNN to solve the problem with pytorch, smth like the models used for imagenet.

Thamks bro, is a great idea

(August 12, 2022, 09:59 AM)Mysterious Wrote: Captcha generally poor implemented like it will be implemented like post request does not send the Captcha, or you can bypass it. Tried that first.

Next thing I recommend  what captcha code (I mean the service like breached also used bb2 captcha) is used or get them, use OCR( you can use the existing code made by someone or make your  own (I will prefer this If you can do it)) and train yourself a model.

right, the website actually seems not so properly written, gonna look into that method. Image processing also looks like another monster to learn lol. Thanks for the suggestion.

If you use OpenCV you can add a thresholding filter to the image to make letters stand out and those thin lines probably can be ignored as they are just too small.

surely they are

There are a large amount of websites which have a captcha but dont actually need it or store the captcha in a session so you can reuse it over and over again.

(August 16, 2022, 04:24 AM)plaga789 Wrote:

(August 11, 2022, 10:43 PM)dthang Wrote: If you start by mapping all the different chars you can find and assigning labels to them, you can use a CNN to solve the problem with pytorch, smth like the models used for imagenet.

Thamks bro, is a great idea

Its a great idea thank you