Captcha Breaking with PHP GD

[tremor77]

that K looks almost like an X. Those are pretty fugly captcha letters, once you get them separated they still look like a pain in the arse. Do you know if this is a standard captcha program or one that's custom made? Like... is it part of a standard CMS?

[TheMindRapist]

Yes here is what the original image looks like.


And thanks for that tip I will mess with it.

You say the background is always the same, I assume that means it always has the same colors, and that those colors are unique. If so, setting to white all pixels with the background colors that do not touch a pixel with a non-background color, then setting to white all pixels that are not a background color then setting to black everything that is not white might give you something clearer to work with.

Edit: Actually, I can't tell if the outline has background colored pixels in it from looking at the image so this may not work.

[adlez]

Just an update on where I'm at.

Using the method suggested above I have been able to get the xy values.

Code: Select all

*60,18
53,19
*161,19
*16,20
158,20
13,21
*134,21
11,22
127,22
10,23
9,24
123,24
8,25
*93,25
7,26
6,27
5,28
4,29
2,34
80,59


The ones with a * next to them are the 5 I need. I'm having difficulty filtering out the rest. I will paste the the method I'm using and continue to debug. If any of you have any suggestions, as always they are greatly appreaciated.

Code: Select all

$coord = array();
for($y=0;$y<$h;$y++) {
  for($x=0;$x<$w;$x++) {
   $rgb = imagecolorat($im, $x, $y);
   $data = imagecolorsforindex($im,$rgb);
   if ($data['red'] == 0){
   ////////////////////////////
    if (!isset($coord[0])){
     $coord[0] = array($x,$y);
     break;
    }else{
     $i = count($coord);
     $coord[$i] = array($x,$y);
     $white = imagecolorallocate($im,255,255,255);
     for ($a=$x;$a<$x+40;$a++){
      for ($b=$y;$b<$y+40;$b++){
       imagesetpixel($im, $a,$b, $white);
      }
     }
    }
    /////////////////////////////
   //once a black pixel is found, make all pixels with a 40x40 radius white//
   }
  }
}


-- Fri Aug 06, 2010 1:46 pm --

Yes here is what the original image looks like.


And thanks for that tip I will mess with it.

You say the background is always the same, I assume that means it always has the same colors, and that those colors are unique. If so, setting to white all pixels with the background colors that do not touch a pixel with a non-background color, then setting to white all pixels that are not a background color then setting to black everything that is not white might give you something clearer to work with.

Edit: Actually, I can't tell if the outline has background colored pixels in it from looking at the image so this may not work.

To clarify, the background for all captchas is that light blue color with the criss cross lines.