Letter I - lower & upper cases causing inappropriate recognition results about captchasolver HOT 9 CLOSED

Works as now all duplicate small "l" (like lucky) and big "I" (like Insane) are not recognized 2 times. If I find some time in the coming weeks & months I might leaverage your incredible captcha creator to consider this challenge riof the right in the YOLO training file.

Thanks again for your outstanding support! Looking forward to follow any of your new projects! ;)

from captchasolver.

Hey,
ty for the hint. Its actually true "i" and "l" are really just the same "image"... carzy, I didn't notice till now :D
But we have no chance to detect what it really is because that are random letters, so I don't know what to do about it. What do you mean by "double recognition"? It should return only one result per letter.
Greetings

from captchasolver.

Thanks for the quick feedback. As I am currently still on vacation not having full access to my virtual machines I will share an example w/ you early next week.

What actually happens when just one "i" or "I" image is screened in the captchas is that the OCR script does call-out the "I" letter twice as from my impressions the OCR does come to the conclusion that both letter cases show a high OCR confidence level.

So my request is not to differentiate between "i" or "I" as this is just impossible (being the same letter case in that font) but to avoid that the one image letter leads to two recognitions of this one letter.

I hope that my explanation does give you more insights into my challenge. As mentioned I will be able to provide an example by next week. ;)

from captchasolver.

I think, @EdgarBlack2 you are saying that an image like "EfGhI" is solved to "EfGhIl", i.e. the image of the last letter "I" is solved to two letters: small "L" and capital "i"?

from captchasolver.

Correct... due to the captcha script limit to 6 letters I have realized that often times the last none-I char is just cut-off leading to captcha failure.

from captchasolver.

Just one adhoc example... the following captcha...

... is resulting into "uXIlmr" using your highly appreciated OCR script. That means the following char gets replaced by another "I" letter as the script get's confused trying to differentiate between lower and upper letter cases of "I". I thought about changing the char range blanking either the lower or upper case "I" char but that destroyed the complete OCR script leading to high failure rates.

Any guidance how to fix this challenge would be highly appreciated. ;)

from captchasolver.

I added this lines into the ocr.js

for(var i = valdResA.length-1; i>=0; i--) { //Remove "I" because big "i" and small "L" -> "l" have the same char in this font
    if(valdResA[i]["c"] == "I") {
        valdResA.splice(i, 1);
    }
}

The proper way would be to train the network again without on of the chars... But this workaround sould do it for now. What do you think?

from captchasolver.

@cracker0dks, thanks for this code snippet. As I do not have full access to my virtual machine being out of town till end of this week for vacation I will test your improved script early next week. I am pretty confident that your code bypass should fix this challenge without having to train YOLO once more. But based on your incredible captcha creator the retraining should be an easy task I will also give it a try in the coming weeks. ;-)

I let you know during next week if the code improvement did fix this challenge.

Thanks again for your outstanding work on GitHub. It is just great how you are sharing your work in full transparency and also granting extremely quick support.

from captchasolver.

Glad to help and that you find it usefull :)
Let me know how it goes 👍

from captchasolver.