How are you supposed to account for noise when doing programming

>>1529173
That's a pretty vague question. What are you trying to do?

File: EfgpcXDUyHw0uLhqRB4afHwVAi8tHh4KgReWjw8FQIvLR6eCoGXFg9PhcBLi4enQuClxcNTIfDS4uGpEHhp8fBUCLy0eHgqBF5aPDwVAi8tHp4KgZcWD0+FwEuLh6dC4KXFw1Mh8NLi4akQeGnx8FQIvLR4eCoEXlo8PBUCLy0engqBlxYPT4XAS4uHp0L4P4yOjD+1H3I5AAAAAElFTkSuQmCC.png (10KB, 286x96px) Image search: [Google]

10KB, 286x96px

Given this problem that needs to be solved, you would need to determine what's noise and what's actual data.
How do you determine what needs to be disregarded?

How are you supposed to account for shitty threads made by pseudo intellectual anally masturbating retards telling themselves they are 'contributing' to the board, that have already been posted to death?
Is it a method of insulting them and determining what kind of threads are seen as significant and insignificant bullshit?

>>1529203
Is that 8gag's captcha? I've noticed that they're not very consistent with the noise; if you reload it enough, you can find an image with little or no added noise.

As a more general statement, you need to figure out how the noise differs from the signal. Maybe the stroke width and/or absence of the second nearby stroke could be used.

The alternative is trying to find a way to recognize the signal that isn't very sensitive to the noise. I think Google had a strong anti-captcha attack that either didn't rely on segmenting the image into letters or tried many possible candidate segmentations.

>>1529173
Just train a neural network and be done with it. That's what all the l33t """"programmers"""" today are doing.

>>1530644
You still have to pick the features to use for input.

Everyone, did you know that someone posted on this thread and then quickly deleted that post?

They did it to put this thread back on page one, but they deleted it for a lot of reasons. This happened all the time, but you should know about it.

I wrote an simple character recognition engine once to solve the captchas for some online poll once. Wasn't able to find the code, though, but from what I remember I extracted a bunch of features from the characters like statistical moments and numbers of holes, and compared them to the usual values for each letter. It was an easy captcha, and I didn't research what the best known techniques for OCR are, so there are probably better ways.

>>1532066
There were some games I played where you had to trace a line and it would grade you based on your accuracy.
I think those used some sort of pixel matching algorithm to determine what lies where.

>>1532494
That's pretty simple to do, you can just calculate the distance from the mouse to the line.

Perhaps relevant: >>1531510
Not sure if correlation would work or if it would be too hard to model all the possible distortions.

File: vew.jpg (297KB, 878x679px) Image search: [Google]

297KB, 878x679px

>>1533361
I was thinking about that and was wondering if I could have a database of alphanumeric characters and perform transformations on them using OpenGL.
The script would go through every possibility and essentially brute force an answer through pixel comparisons of a separated character against a transformed template character.

The GPU is optimized for matrix calculations so I'd wonder if it could do a highly accurate and speedy computation by this.

>>1529203 looks like a sine wave, so it might be possible to detect the parameters of the sine wave and reverse the distortion.