Blue Eye Macro

Hello, I can't get my tessaract to read the captcha.

How can I update, train or install another language for this captcha?

It's only contains number 1, 2, 3 and 4. But it's not a ble to read

If it's a captcha you're trying to read, you may be out of luck. The whole point of captcha is that OCR doesn't work on it.

Thank you for your reply

Tessaract seems to work 100% fine with this captcha when I remove lines and so on using paint.

My question is,

1.How can I make this program automatically removes the lines and then read the number?

2.If I understand it right, we can upgrade(train?)tessaract and create a new language (this captcha) using edited version of the captcha to get the original texts from the picture? So it will be able to read this captcha






I believe that it's possible that we can bypass captcha using OCR, because this guy have already done it.

https://www.youtube.com/watch?v=oYN5eyLFaJA

But the code seems very difficult and I couldn't follow him, so I have to ask you the alternative ways or how do I remove those RGB from your Profiler

- This captcha only contains number 1,2,3,4 it's from Legend of Edda Online

These are the examples.









I'm very thankful that you created this plugin, it's very helpful and makes me very close to goal.

Thankyou.

I haven't trained tesseract before, this site may help
https://code.google.com/p/tesseract-ocr/wiki/TrainingTesseract3

it's possible since it's a very specific line pattern, you might be able to write some BEM code to read the colors that occur most/least and filter out/for those colors using the plugin call. Assuming of course the pics you showed were filtered and started in different colors. It would take some math for sure. It might also be possible to scan vertically for how many black pixels are present and drop pixels in columns that don't pass a threshold. In other words, I don't see an easy way to do it, but it should be possible if you're creative

Thank you for you suggestion, I've already came up with the idea.

I noticed that the numbers that contain black color use different RGB from the line pattern, I've spent over 9 hours collecting RGB of those line patterns.

If you can create function to ignore those RGB or delete those RGB from the screenshot, it would only be left with pure numbers!!!

Here are the RGB that must be removed in order to get Tessaract to be able to read the captcha.

Line patterns of the captcha
0 0 0
0 4 0
8 8 8
8 12 8
16 16 16
24 24 24
33 32 33
33 36 33
41 20 41
41 40 41
49 48 49
49 52 49
49 48 49
49 49 49
57 60 57
58 57 58
58 61 58
66 65 66
66 69 66
66 69 74
74 77 74
82 81 74
82 81 82
82 81 87
82 85 90
82 89 90
90 89 90
99 97 99
99 101 99
99 101 99
107 105 107
107 109 107
115 113 115
115 113 123
115 117 123
115 117 115
123 121 115
123 121 123
123 125 123
132 130 132
132 134 132
140 138 140
140 142 140
148 146 148
148 150 148
156 154 156
156 158 156
164 166 164
165 162 165
165 166 165
173 170 173
173 174 173
181 174 173
181 178 181
181 182 181
189 186 189
189 190 189
197 194 197
197 198197
198 199 198
206 202 206
206 203 206
206 206 206
206 207 206
214 211 214
214 215 214
222 219 222
222 223 222
230 231 230
231 231 231
239 235 239
239 239 239
247 243 247
247 247 247
255 251 255
255 255 255

My question is how can I make this program to ignore or delete these specific RGB?

Here is how it should look like after all these RGB are removed from the picture





and also, to be much much more accurate. How do I force tessaract's output equals to 4 digit numbers, no more or less. And how do I force it to generate only 1,2,3,4?

Thank you.

I'd like to help, what you're asking appears to require some coding. With school I'm pretty busy at the moment. I'll see what I can do this weekend but doubt I'll have anything before then. Two of my computers have failed hard drives so I'll have to see if my plugin source is still current. If it's not I may have to make a new image filter plugin. Of course, there are always more ways to get things done. have you tried reducing the block into four blocks, one for each character? You might be able to write some BEM logic to figure out where the midpoints between characters are and split the images between those and limit the output to one character in the plugin. You might be able to zoom out in tesseract to the point it ignores the excess lines if there is only one number per image.

I apologize but the reason I'm looking at this process is the reason I need it. I have a severe Sequencing disorder and I have a very hard time, often an impossibly hard time tracking or even recognizing things going on in various games that I try to play. Is this working to a level it can be used to convert images inside games to text that can be read or that can create alerts that let me know something is happening?

I'm trying to follow and I'm not sure I understand what stage this is at and if it can be used in a game or other settings yet?

Thank you all so much, this is an awesome thing you are doing.

Absolutely, it works best when the area and color of text you're looking for is known.

You'll have to write the logic for the alert etc, but it works, give it a try!