phpmyfaq 160

[Thorsten]

Hi,

I received no mail....

bye
Thorsten

[jazcyk]

the validation performed against tha captcha code is case insensitive

Then I don't understand why 'M' worked in some situations and 'm' did not. Similar with 'O' and 'o' (as far as I remember).

Also (If possible)
*always use serifs with 'I' to distinguish from 'l' and best avoild 'l'
* maybe avoid 0 (zero) and O? (as they will never be unambigious)
* problem with '1' is that if turns clockwise, eyes sees it as '7'. Always put a serif on the 'foot' of '1' if possible.
* fix the issue that 'B' displays as 'a' (upper half of vertical bar is cut off).
* two of the graphics that I sent has a strange glyph, that I simply cannot associate with any latin letter. Find out what it is - fix it or remove it!
* turn a little down for background variation. Avoid too dark colors here.
* If possible avoid changing position and rotation too much at the same time for a character.

That should be 98%

Where does it get its bitmaps from? If is is an installed font that is used then always use serif fonts (Times etc.). They will be more unique and recognizable even if rotated - not the case with sans-serifs from my experience.

[Thorsten]

Hi,

you can add or remove the font files in the folder font/ as you want.

bye
Thorsten

[jazcyk]

sent again to thorstenATphpmyfaqDOTde
.. this is what the mail-link of your member info says.
The first mail was not returned!

???????

[matteo]

the validation performed against tha captcha code is case insensitive

Then I don't understand why 'M' worked in some situations and 'm' did not. Similar with 'O' and 'o' (as far as I remember).

Hi jazcyk,
I'd suggest you to keep the faqcaptcha table opened with data ordered by captcha_time descendant: this is the way I'm testing the captcha class far from when it was QAed before releasing it for PMF 1.6.0 alpha. I think this way of testing will help you in your great work of understanding the "failures".
I think you'll agree with us that, on average, we've reached a good trade-off for distortion and there is room for improving it in some next release, as Thorsten said

[Thorsten]

Hi,

I received your mail... I think, the images are quite hard but readable...

bye
Thorsten

[jazcyk]

Thanks for the tip on how to use the table.

Well .. about the 'balance': probably someone will complain whatever you do . We only have a few user inputs so it is no deal for us. But personally I think they are too hard to solve. around 1/3 made me uncertain and more than 5% were not possible.

[jazcyk]

Code: Select all

... I think, the images are quite hard but readable...

Well .. how do you read them. I did not manage to solve those three!

[Thorsten]

Hi,

I think these are:

1. CRGHIA

2. OQIXWW

3. RAPQHI

bye
Thorsten

[jazcyk]

Impossible to verify now. But you are right about the 'A'.
I found out in between.

I just installed the RC1 on my test mirror, and everything works as expected.

[matteo]

Impossible to verify now. But you are right about the 'A'.
I found out in between.

As Thorsten said, you can also use different TTF fonts: the captcha code image is built using a different font (random choice through the PATH_TO_PMFINSTALL/font/*.ttf files) for each letter. E.g., try to rename all the .ttf as .ttf_ except for captcha_2.ttf, you'll clearly see some benefits but... the clearer the lesser the captcha will be usefull for spam protection.

I just installed the RC1 on my test mirror, and everything works as expected.

Glad to read it.

[jazcyk]

I tried renaming all three fonts and experimenting with others. It is correct what I thought and wrote that SERIF fonts work better than SANS-SERIFS. With SERIFS it seems that ambiguities are more rare.

The 'Bitstream Vera Serif (Bold or not)' is a good example. I think it is availabale as it is distributed with Open Office and is included with any Linux distribution that I know. And it is small in size too (58 KB) though some glyphs are mapped that we don't use - such as Greek letters . The copyright notice in the font allow for embedding and permanent install.

As any Graphics specialist know SERIFs are designed to 'catch the eye'. That is why they are mostly use for newspaper headings!

And I still request that 'O' (letter) and '0' (zero) are considered identical by the code

[matteo]

The 'Bitstream Vera Serif (Bold or not)' is a good example. I think it is availabale as it is distributed with Open Office and is included with any Linux distribution that I know. And it is small in size too (58 KB) though some glyphs are mapped that we don't use - such as Greek letters . The copyright notice in the font allow for embedding and permanent install.

Thanks for the hint on that particular font, we'll double check for the license issue and (maybe) insert it as another option among the other fonts.
We know that serif fonts are preferred for quality reading but... our main goal is to provide a good feature (the captchas) for the main stream, and for good I mean, in this case, giving each PMF administrator the way to choose those fonts that will provide them a good trade-off for best human captcha code reading.
We'll also add some comments on the code to focus on their effect.

And I still request that 'O' (letter) and '0' (zero) are considered identical by the code

We'll implement this request. Please, stay tuned for its availability.
Thanks for your testing.
Ciao,
Matteo

[jazcyk]

I can see in the devblog that you did this now.
What files should I replace?

Thorsten writes something like 'if this will be misused .. we will remove it'. I don't understand how it can be 'misused'. My point is that a user has no chance to know if it is an '0' or an 'O' that is displayed. User can't know either if the use of '0' (zero) is disabled. If a captcha is for instance 'V15O7A' there is much chance that 'O' (letter) is interpreted as '0' (zero) by user as it occurs among other numbers.

The problems with identifying '1', '7', 'I', and 'L' can be solved by using appropriate fonts - but that won't ever solve the 'O' <> '0' issue.

[david_de]

The problems with identifying '1', '7', 'I', and 'L' can be solved by using appropriate fonts - but that won't ever solve the 'O' <> '0' issue.

What about a slash zero? Would that help to solve the problem? I have not had any problems so far in my limited testing of 160 RC but I did in the beta 160. I can only assume I am either very lucky now or something changed.

Without something to distinguish the only option I see is to eleminate the problem characters. I suspect a lot of humans that fail the test a time or two would move on. Especialy if it is their first attempt to post.

Jazcyk solution to make 0 (zero) and o (oh) the same in code might be the easiest.

Best solution to a lot of problems would be to shoot all spammers then hang them.