Server Help

[Solo Ace]

I think it's interesting how some software can determine what the digits are in CAPTCHA images.
I started working on decoding some easy CAPTCHAs too.

My school's intranet site's security was "increased" by another student from another part of the country (uhm, why is it an intranet site when it allows external logins? Isn't that an extranet?)
He's even getting paid for it.
His PHP code is generating this poor captcha image:



Before this I've never done any image programming, except for a little GD2 with PHP, and I'm not going to read any manuals/tutorials, I'm fine on my own it seems.

I'm just reading the image and doing a (pseudocode):

Code: Show/Hide if ((R == 255 && G == 204 && B == 255) || // background    (R == 69 && G == 58 && B == 40) || //lines    (R < 250 && G > 10 && B > 10))    R =  G = B = 255

What we have left is easily readable by an OCR application, like GOCR for example.



I convert the bitmap to a PPM (otherwise GOCR will barf), run "gocr.exe intranet2.ppm > intranet2.txt" and find "JeDADck" in intranet2.txt.

I think it took me like an hour (yeah I know it's long) to figure everything out, but I think it's fun.

[Solo Ace]

[SpecShip]

You enjoy terrorizing Cerium (dailup user)?

I don't care what's your excuse, posting bmps on forums should get you hanged unless it's a 3kb file sized 30x30 or something...
Or a SS banner.


ROFL @ kings game...browser games are for gay handicapped people.
_________________
Replacing yazour untill the whore returns.

"I could run a ss server on my car stereo!" -Xalimar
"Liberta tuit ma ex infernis" -Event Horizon
"I know too much about nothing." - Mine GO BOOM
"Hmm anyway, back to my kingdom hearts." - Chambahs

[Solo Ace]

Uh, if I'd compress the data the results of what I've done to the images would be a little vague, so I kept the original images.
There's no point in trying to make a car faster/leighter by removing the engine or whatever keeps the car going.

I'm working on Age 6 now, which seems to be even easier.

[Mine GO BOOM]

[Smong]

What is Age 5/6? Is that a game or a captcha system? Also with ten/fourteen you didn't explain how you removed the lines even though they are the same color as the text.

[Solo Ace]

Fine MGB, I realized they'd still be identical, but whatever, I still wanted to keep what I worked with.

I don't really know what the age 5/6 are.
Uh it's just... a time, I guess.

[Bak]

looks like you could attempt to make the smallest rectanagle possible around the text then rotate it so the the rectangle is parallel the the bottom of the image... then try the OCR. Also looks like they're using dashes you might want to scan for those
_________________
SubSpace Discretion: A Third Generation SubSpace Client

[Solo Ace]

That's exactly what I wanted to do, but it'd be useless now.

I'm working on the "Age 6" CAPTCHA, which looks like this:


I have several ideas for this one already, I think I'm pretty close to filling the digits completely and I can easily get rid of the pixels unrelated to the digits.

[Quan Chi2]

Wow, thanks Solo. I always wondered what to search for to find out what those images were and how they are generated. Im reading some articles on google about it now.

[Solo Ace]

A CAPTCHA (an acronym for "completely automated public Turing test to tell computers and humans apart", trademarked by Carnegie Mellon University) is a type of challenge-response test used in computing to determine whether or not the user is human.

Well, I just want to be able to break some, I won't abuse it when I can.
I'm about to show my school's principal a brute-force of his password for his account, though, but he challenged me.

[Quan Chi2]

Don't do that. I told the technology admin that the security was weak - both firewalls and filters. He banned me from the school network the second the filters shut down without a trace.. -.-

So I wouldn't open my mouth unless it was urgent. Just act like you don't know how to do it when you confront your principal. lol

[SpecShip]

Well, I have no idea how to put it code-wise, but logic wise, it would seem to me that the key is to query for distance between the dots then lock down on clusters whereas the density is high.

[Quan Chi2]

What?

[SpecShip]

My apologies Quan.
My messages carry a hidden requirement of ReaderIQ>100 inorder to understand.

[Mine GO BOOM]

[Mine GO BOOM]

And so people actually see it, here is the source code/executable for the program that did the above. Either run captcha.exe, or pass it parameters in the format of <imagename> <distance> <threshold>

Captcha.zip

Uses Corona as an image loading/saving format, because its my favorite for simple pixel control and loads a ton of different image formats.

[Quan Chi2]

Amazing.. lol

[Solo Ace]

Nice MGB. I'm stuck now haha.

I've tried some things, but I'm an idiot, it didn't work well. Well, not "well", it didn't work at all.
My best attempt even killed GOCR.

[Bak]

shurnk it down to 25% size (interpolate each point from every four), filled completely black with yellow, then expanded it back 400% to normal size... looks like that will isolate the numbers:

[Mine GO BOOM]

A simple fix. Right now, the program only will check pixels that are already active. Just remove if (bitmap[GetXY(col, row, cols)]) statement, and change the color outputs so they are inverted. The if(count) goes to black (0) now, while the else goes to white (255).

Running with the radius of 15 and threshold of 15 comes out with:


Run that through gocr -v 7 -c _ -C "1234567890" captcha-015-015.png

Code: Show/Hide # Optical Character Recognition --- gocr 0.40 # options are: -l 0 -s 0 -v 7 -c _ -m 0 -d -1 -n 0 captcha-015-015.png # using unicode # popen( pngtopnm captcha-015-015.png ) # PNM P6 x=360 y=220 c=255 head=-1 # db_path= (null) # OTSU: thresholdValue = 0 gmin=0 gmax=255 # scanning boxes 6 # auto dust size = 2 (mX=28,mY=37) # remove dust of size  2 histo=3,0(?=0),0(?=0),...   3 cluster removed #   4 white pixels removed, cs=160 # smooth big chars 7x16 cs=160 ... 122 changes in 3 of 3 # detect barcode , 0 bars, boxes-0=3 # detect pictures, frames, noAlphas, mXmY= 55 73 ...  0 - boxes 3 # averages: mXmY= 55 73 nC= 3 n= 3 # remove boxes on border pictures= 0  rest= 3  boxes?= 3 deleted= 0, within pictures  pictures= 0  rest= 3  boxes?= 3 . deleted= 0, pictures= 0  rest= 3  boxes?= 3 # rotation angle (x,y,num) (100352,5120,1) (0,0,0), pass 1 # rotation angle (x,y,num) (100352,5120,1) (0,0,0), pass 2 # detect longest line - at y=0 crosses=  0 my=0 - at crosses=  0 dy=0 # scanning lines # trouble on line 1: #  bounds: m1=  46 m2=  11 m3= 104 m4= 104  my=  73 #  counts: i1=   1 i2=   1 i3=   0 i4=   2 #  all boxes of same high! - lines= 1 # add line infos to boxes ... done # divide vertical glued boxes, numC 3 # searching melted serifs ...   0 cluster corrected, 0 new boxes # glue broken chars ...   0 times glued, remaining boxes 3 # detect dust2, ...    1 +   0 boxes deleted, numC= 2 # check for word pitch ... min=28 max=28 pitch_p=29 #  ...  no spaces found # step 1: char recognition unknown= 2 picts= 0 boxes= 2, 1 of 2 chars unidentified # debug: unknown= 1 picts= 0 boxes= 2 # step 2: try to compare unknown with known chars - found 0 # step 3: try to divide unknown chars, numC 2 # list shape   0 x=  80   46 d= 68 105 h=1 o=1 dots=0 e000 (?) # list box dots=0 c=(?) ac=(?) mod=(0x00) line=1 m= 0 10 104 114 r= 35 0 # list pattern   x=  80   46 d= 68 105 t=1 3 ...................................@@@..............................<- ...............................@@@@@@@@@.@@@@....................... .............................@@@@@@@@@@@@@@@@....................... .........................@@@@@@@@@@@@@@@@@@@@@@@@@.................. ......@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@................. ....@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@................ ....@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@................. ...@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@................ ...@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@................. ....@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@.................. .....@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@................... ...........@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@.................. ...................@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@.................. ....................@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@................. .......................@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@............... .......................@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@............... .....................@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@............... ......................@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@................ ......................@@@@@@@@@@@@@@@@@@@@@@@@@@@@@................. .....................@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@................. .....................@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@................. ......................@@@@@@@@@@@@@@@@@@@@@@@@@@@@@................. ......................@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@................ .........................@@@@@@@@@@@@@@@@@@@@@@@@@@................. ........................@@@@@@@@@@@@@@@@@@@@@@@@@@@................. ......................@@@@@@@@@@@@@@@@@@@@@@@..@@................... .......................@@@@@@@@@@@@@@@@@@@@@@....................... ..................@@@@@@@@@@@@@@@@@@@@@@@@@@@....................... ......@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@.........@...... ......@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@.... @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@.. @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ ..@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@.. ....@@@......@@@..............@@@@@@@@@@@@@@..........@@@@@@@@@..... # set space width to 29 # insert space between words (dy=115) ... found 0 # step 4: context correction Il1 0O # store boxtree to lines ...get_least_line_indent: page_width 360, dy 0 Line 1,  y 46, raw indent 80, adjusted indent 80 Minimum indent is 80 ... 1 lines, boxes= 1, chars= 1 # debug: (_)= 1 picts= 0 chars= 1 (4)=1 Elapsed time: 0:01:19.787.

Well, has a problem with the one, not surprising. Reduce the image by a forth, get the following:

gocr -v 7 -c _ -C "1234567890" captcha-015-015-sm.png

Code: Show/Hide # Optical Character Recognition --- gocr 0.40 # options are: -l 0 -s 0 -v 7 -c _ -m 0 -d -1 -n 0 captcha-015-015-sm.png # using unicode # popen( pngtopnm captcha-015-015-sm.png ) # PNM P6 x=90 y=55 c=255 head=-1 # db_path= (null) # OTSU: thresholdValue = 0 gmin=0 gmax=255 # scanning boxes 3 # auto dust size = 1 (mX=13,mY=18) # remove dust of size  1 histo=0,0(?=0),0(?=0),...   0 cluster removed #   0 white pixels removed, cs=160 # smooth big chars 7x16 cs=160 ...  16 changes in 2 of 3 # detect barcode , 0 bars, boxes-0=3 # detect pictures, frames, noAlphas, mXmY= 13 18 ...  0 - boxes 3 # averages: mXmY= 13 18 nC= 3 n= 3 # remove boxes on border pictures= 0  rest= 3  boxes?= 3 deleted= 0, within pictures  pictures= 0  rest= 3  boxes?= 3 . deleted= 0, pictures= 0  rest= 3  boxes?= 3 # rotation angle (x,y,num) (25600,1024,1) (0,0,0), pass 1 # rotation angle (x,y,num) (25600,1024,1) (0,0,0), pass 2 # detect longest line - at y=0 crosses=  0 my=0 - at crosses=  0 dy=0 # scanning lines # trouble on line 1: #  bounds: m1=  12 m2=   2 m3=  24 m4=  24  my=  18 #  counts: i1=   1 i2=   1 i3=   0 i4=   2 #  all boxes of same high! - lines= 1 # add line infos to boxes ... done # divide vertical glued boxes, numC 3 # searching melted serifs ...   0 cluster corrected, 0 new boxes # glue broken chars ...   0 times glued, remaining boxes 3 # detect dust2, ...    1 +   0 boxes deleted, numC= 2 # check for word pitch ... min=9 max=9 pitch_p=10 #  ...  min=25 max=25 v=0.000000 mono=1 pitch_m=25 # step 1: char recognition unknown= 2 picts= 0 boxes= 2, 0 of 2 chars unidentified # debug: unknown= 0 picts= 0 boxes= 2 # step 2: try to compare unknown with known chars - found 0 # step 3: try to divide unknown chars, numC 2 # set space width to 25 # insert space between words (dy=28) ... found 0 # step 4: context correction Il1 0O # store boxtree to lines ...get_least_line_indent: page_width 90, dy 0 Line 1,  y 12, raw indent 20, adjusted indent 20 Minimum indent is 20 ... 1 lines, boxes= 2, chars= 2 # debug: (_)= 0 picts= 0 chars= 2 (1)=1 (4)=1 Elapsed time: 0:00:42.897.

Returns 14. Image defeated. Now try it with a couple more sample images, and maybe even add the reducing to the code itself to make the image smaller.

[Solo Ace]

Bah, I feel dumb now.
Maybe programming isn't meant for me, or maybe I just need more practice or intelligence.

Have you compiled the GOCR yourself?

gocr -v 7 -c _ -C "1234567890ero" captcha-017-017.png gave me:

Code: Show/Hide # Optical Character Recognition --- gocr 0.40 # options are: -l 0 -s 0 -v 7 -c _ -m 0 -d -1 -n 0 captcha-017-017.png # popen( pngtopnm captcha-017-017.png ) ERROR src\pnm.c L208: sorry, compile with HAVE_POPEN to use pipes

Note: I'm using 'ero' in "1234567890ero" because I've seen an image containing "err" once, guess that's a good reason to check for it.

I'm using

Code: Show/Hide FIBITMAP *image = FreeImage_Load(FIF_PNG, filename, 0); image = FreeImage_Rescale(image, cols / 4, rows / 4, FILTER_BICUBIC); FreeImage_Save(FIF_BMP, image, "temp.bmp", 0); // this one's just so I can easily SEE the resized image :P FreeImage_Save(FIF_PPM, image, "temp.ppm", 0); FreeImage_Unload(image);

For some reason I only get the correct result '14' when using distance and threshold 13. Otherwise I'm getting _4 or 29.
Could this be because it's saving it as ppm?
I chose the bicubic filter for rescaling it, it's a lot smoother than bilinear, but I don't know if that really is an advantage here.

This is the code I have now.

Code: Show/Hide #include <stdio.h> #include <stdlib.h> #include <string.h> #include "corona.h" #include "FreeImage.h" #define IMAGE_FORMAT corona::PF_R8G8B8 /* RGB mode - 8 bits each */ #define GetXY(x,y, w)  ((x) + ((w) * (y))) #define MIN(a, b) ((a) < (b) ? (a) : (b)) #define MAX(a, b) ((a) > (b) ? (a) : (b)) #define SQ(a) ((a) * (a)) #define DISTANCE(a, b, c, d) (SQ(a - c) + SQ(b - d)) void savepixels(const char *filename, int width, int height, unsigned char *bitmap) {    unsigned char *pixels = new unsigned char[width * height * 3];    unsigned char *p = pixels, *b = bitmap;    int col, row;    for (row = 0; row < height; row++)    for (col = 0; col < width; col++)    {       *p++ = *b;       *p++ = *b;       *p++ = *b++;    }    corona::Image *img = corona::CreateImage(width, height, IMAGE_FORMAT, pixels);    corona::SaveImage(filename, corona::FF_AUTODETECT, img);    delete img; } int main(int argc, char **argv) {    corona::Image *img;    unsigned char *pixels;    char filename[255], *nl;    int threshold = 0;    int distance = 0;    int pixel;    char str[255];    int rows, cols;    int row, col;    unsigned char *bitmap, *p;    unsigned char *outmap;    FIBITMAP *image;    if (argc > 1)    {       strncpy(filename, argv[1], sizeof(filename) - 13);       filename[sizeof(filename)-13] = 0;       if (!strcmpi(filename, "/?") || !strcmpi(filename, "-h") || !strcmpi(filename, "--help"))       {          printf("captcha <filename> <distance> <threshold>\n");          return 0;       }              if (argc > 3)       {          distance = atoi(argv[2]);          threshold = atoi(argv[3]);       }    }    else    {       printf("Enter captcha filename: ");       fgets(filename, sizeof(filename) - 13, stdin);       nl = strchr(filename, '\n');       if (nl)          *nl = 0;    }    img = corona::OpenImage(filename, IMAGE_FORMAT);    if (!img)    {       printf("Could not open file %s\n", filename);       return 1;    }    pixels = (unsigned char*)img->getPixels();    rows = img->getHeight();    cols = img->getWidth();    bitmap = new unsigned char[rows * cols];    p = bitmap;    outmap = new unsigned char[rows * cols];    //convert to grayscale of a single byte    for (row = 0; row < rows; row++)    for (col = 0; col < cols; col++)    {       pixel = *pixels++;       pixel += *pixels++;       pixel += *pixels++;       *p++ = pixel / 3;    }    //free corona loading    delete img;    while (distance <= 0 || distance >= 1000)    {       printf("Enter distance to check: ");       fgets(str, sizeof(str) - 1, stdin);       distance = atoi(str);    }    while (threshold <= 0 || threshold >= 255)    {       printf("Enter threshold: ");       fgets(str, sizeof(str) - 1, stdin);       threshold = atoi(str);    }    //check our threshold    for (row = 0; row < rows; row++)    for (col = 0; col < cols; col++)    { //      if (bitmap[GetXY(col, row, cols)])       {          int count = 0;          int x, y;          int dhalf = distance / 2 + 1;          //could optimize here heavily, by only checking inside a circle rather than square + distance          for (x = MAX(col - dhalf, 0); x < MIN(col + dhalf, cols); x++)          for (y = MAX(row - dhalf, 0); y < MIN(row + dhalf, rows); y++)          {             if (SQ(distance) > DISTANCE(col, row, x, y) && bitmap[GetXY(x, y, cols)])                count++;          }          if (count >= threshold)             outmap[GetXY(col, row, cols)] = 0;          else             outmap[GetXY(col, row, cols)] = 255;       } //      else //         outmap[GetXY(col, row, cols)] = 0;    }    //save output    nl = strrchr(filename, '.');    if (!nl)       nl = &filename[strlen(filename) - 1];    sprintf(nl, "-%03d-%03d.png", distance, threshold);    savepixels(filename, cols, rows, outmap);    image = FreeImage_Load(FIF_PNG, filename, 0);    image = FreeImage_Rescale(image, cols / 4, rows / 4, FILTER_BICUBIC);    FreeImage_Save(FIF_BMP, image, "temp.bmp", 0); // this one's just so I can easily SEE the image :P    FreeImage_Save(FIF_PPM, image, "temp.ppm", 0);    FreeImage_Unload(image);    system("gocr -v 7 -c _ -C \"1234567890ero\" temp.ppm > result.txt 2>&1");    delete bitmap;    delete outmap;    return 0; }

I'll do a few test runs on other images and post the results.

Oh and as you can see I'm using the FreeImage library, I don't know if Corona is capable of what I used FreeImage for, but whatever, that's not a problem for now.

[SpecShip]

Must restrain myself...must resist urge...

[Mine GO BOOM]

[t3rmin4t0r]

Hi,
Sorry i am new here. I want to convert the following image (captcha) in to text.. I am using vb6 . How to do it? Would you like to share its coding?



I can convert its background into black color.. But wats next?

-Thanks in advance