Registration

[EoT]

Have you considered putting in a image option where you need to copy the letters in the image to the text box in order to get to the next page? because this site gets alot of advertisements and it would be better if you implemented this or perhaps a e-mail verification.

[Exodus]

Email validation would be nice. This site does get too many bots. Of course I registered a while ago so there could be validation for all I know.

[Obsidian Tides]

Jay was supposed to have a hack for that, but he's obviously swamped, so there's no telling when we were going to get that.

[YopY]

The image verification business is (should be) installed by default with a load of vB versions, including the current one. Might just be disabled, or you just didn't see it (if you did). But nowadays image verification won't work anymore, as shown by this page. It and probably the others use neural networks to recognize letters and such on pretty much every word, which is a funky way of explaining a program that can differentiate characters based on some elementary mathematics and a lot of users telling it which character is what. I made a simple one as well in about half an hour, it could keep two letters apart basically.

How it works? First off, you have a field of squares (pixels, in this case). A pixel can either be on (1) or off (0).

Behind each pixel is a numerical value, starting at 0. Finally, you unleash a fairly simple mathematical function on it, which does something like

result = (grid space 1's value * grid space 1's on/off state (1)) + (grid space 2's value * grid space 2's on/off state) + ... etc

So if you have a grid, all it does is calculate the sum of the on square's values. Great.

Now, in a single-neuron neural network, all it can do is say yes or no, or result is higher then 0, or below 0 in this case. So, you fill in a letter, say an A, as a 100*100 image with a big black A on it. The script does the summing, determines the sum of all black pixels, and says 0, it's a B (for example).

This is where the user comes in. He/she says WRONG, NUBL OOOLOLO and tells the computer so by pressing a button. The computer goes to hide in shame in a corner, and goes calculating again.

It has to update the values now. I forgot the exact formula, but it does something like add the amount it's off with the result (1 in this case, can be more later on) (can be positive or negative) with each filled in square's value. Thus, when you re-run the method again, the outcome is higher or lower with the same character, and now it's properly seen as an A.

Do this plenty of times with a gazillion variations of the letter A, and it can discern As from Bs like wtfbbq. Mine only took a handful of tries to discern an A from a B, although that was a fairly small grid. But the principle is the same. Basically, it's a learning system, and thus it can discern anything from everything. The one used in the link above is just a lot more advanced, we'll get that next year <3.


/emo rant.

[EoT]

Originally Posted by YopY

The image verification business is (should be) installed by default with a load of vB versions, including the current one. Might just be disabled, or you just didn't see it (if you did). But nowadays image verification won't work anymore, as shown by this page. It and probably the others use neural networks to recognize letters and such on pretty much every word, which is a funky way of explaining a program that can differentiate characters based on some elementary mathematics and a lot of users telling it which character is what. I made a simple one as well in about half an hour, it could keep two letters apart basically.

How it works? First off, you have a field of squares (pixels, in this case). A pixel can either be on (1) or off (0).

Behind each pixel is a numerical value, starting at 0. Finally, you unleash a fairly simple mathematical function on it, which does something like

result = (grid space 1's value * grid space 1's on/off state (1)) + (grid space 2's value * grid space 2's on/off state) + ... etc

So if you have a grid, all it does is calculate the sum of the on square's values. Great.

Now, in a single-neuron neural network, all it can do is say yes or no, or result is higher then 0, or below 0 in this case. So, you fill in a letter, say an A, as a 100*100 image with a big black A on it. The script does the summing, determines the sum of all black pixels, and says 0, it's a B (for example).

This is where the user comes in. He/she says WRONG, NUBL OOOLOLO and tells the computer so by pressing a button. The computer goes to hide in shame in a corner, and goes calculating again.

It has to update the values now. I forgot the exact formula, but it does something like add the amount it's off with the result (1 in this case, can be more later on) (can be positive or negative) with each filled in square's value. Thus, when you re-run the method again, the outcome is higher or lower with the same character, and now it's properly seen as an A.

Do this plenty of times with a gazillion variations of the letter A, and it can discern As from Bs like wtfbbq. Mine only took a handful of tries to discern an A from a B, although that was a fairly small grid. But the principle is the same. Basically, it's a learning system, and thus it can discern anything from everything. The one used in the link above is just a lot more advanced, we'll get that next year <3.


/emo rant.

Just wow.. then perhaps the e-mail verification is the way to go? hahaha

[Daryl Falchion]

^That's what I was thinking or a Moderation Queue. However both are a pain the ass as the mods have got to regularly check them and if there's not one person assigned either everyone does it or nobody does.
Also, moderation or email verification runs the risk of alienating potential members because some people just won't hang around long enough to bother.

This has been a totally unhelpful post I know.