pixelsearch help

[brodie28]

I have a series of images on my harddrive and I need to know how I would go about writing a program to do a pixel search of the screen and see if any of the images were found.

I need this to get past the visual confirmations on login pages and such. The site I need it for does not use random colors or backgrounds so I think it should be possible. It needs to search the page for all 27 images (A-Z and 0-9) and when they are found output in order from left to write so that I have the security code.

[clearguy]

Look at your helpfile about

PixelGetColor ( x , y )
PixelSearch ( left, top, right, bottom, color [, shade-variation] [, step]] )

[brodie28]

I have read it but I do not think it provides any real idea of how to achieve what I am trying to.

I will re-read it when I get home in an hour...

So what I need to do is easily accomplished?

[clearguy]

Use pixelgetcolor,to get the needed color,do this for each file...use autoit window info for the coordinates

Then put pixelsearch on your script(if pixel not found @error = 1)...any questions?

[brodie28]

And this will be able to find and tell the difference between different letters and numbers, and list them from left to right?

[Paulie]

  brodie28 said:

And this will be able to find and tell the difference between different letters and numbers, and list them from left to right?

it involves doing MANY 'pixelgetcolor's

you know that the image of an 's' has a black pixel at 200, 200

well you go

$color = PixelGetColor(200,200,)
If Hex($Color) = 000000 then
Msgbox(0, "Result", "could be an 's'.")
Else
Msgbox(0, "Result", "could not be an 's'.")
Endif

[brodie28]

So that is the only way to do what I need?

It would be extremely tedious to do this for 36 different characters....

[Paulie]

  brodie28 said:

So that is the only way to do what I need?

It would be extremely tedious to do this for 36 different characters....

Exactly why OCRs are not fun to make

[brodie28]

Is there no way to just scan all of the pixels of the image files, and just compare it with what is on the screen?

[brodie28]

The thing I don't get is that pixelsearch searches for one color only... I need it to search for a pattern of pixels.

[flyingboz]

There are a few toolkits out there... As someone mentioned, OCR is programatically hard -- you have to try to mimic how your brain listens to your eyes. This is also computationally expensive.

Given that so many sites are using these things expressly to *block* automation should indicate to you that it is a non-trivial problem to solve. People don't pay programmers to stop these things for nothing.

That being said, there are some toolkits out there that, should you wish to purchase them, may provide assistance in you being able to build some logic that can test for your images.

Have you explored other options? Were they careful enough to keep the file sizes the same? All files have the same background? All files have the same name in the html source? Presumably they didn't leave any stupid holes open like that.

Alternatively, you can start your own high volume website, and pay people to solve the things for you in order to view your content. (this is what the spammers and robot creators do -- hijack the image, put it on a porn site, and get the slavering masses to decode the text for them.)

Edited August 21, 2006 by flyingboz

[brodie28]

All the images are the same file size, file name and html code. So no obvious loops.

Am I to understand that autoit cannot reasonably code around this problem?

You did give me an idea about downloading the picture and having someone else work it out though.

[buzz44]

http://www.autoitscript.com/forum/index.ph...c=8442&st=0

http://www.autoitscript.com/forum/index.ph...=20251&st=0

http://www.autoitscript.com/forum/index.ph...7807&hl=OCR

http://www.autoitscript.com/forum/index.ph...=15349&st=0

http://www.autoitscript.com/forum/index.ph...8404&hl=OCR

http://www.autoitscript.com/forum/index.ph...1231&hl=OCR

[brodie28]

I have looked at all of those...

What I am having trouble with are the following areas:

How do I save character definitions... Let's assume that black is equal to 1 and white is equal to 0. The number seven could look like this:

00000000

01111100

00001100

00001100

00001100

How do I save this in my script (and of course my script will need to be able to do it with more than two colors.

And how do I ask autoit to search for a complex pattern of pixels on the screen like the one above?

[brodie28]

bump

[SmOke_N]

  brodie28 said:

How do I save this in my script (and of course my script will need to be able to do it with more than two colors.

And how do I ask autoit to search for a complex pattern of pixels on the screen like the one above?

By their checksum values.

[brodie28]

So I can get their checksum values, and then search a rectangle on the screen for them?

How do I search for checksum values?

[AgentSmith15]

I'm guessing you don't read very much.Instead of re-inventing the wheel there is already stuff out there that cracks OCR.