Believe nothing of what you see...

netmaestro · Post by **netmaestro** » Thu Sep 15, 2005 2:47 pm

Well, with the mask off I'm getting widely differing readings. It complicates things a bit when the two crosspieces are on different angles with shadows applied, but if I eyedroppper the similarly-lit areas of the two crosspieces I can't find one spot that's the same. This is the screenshot I took (mask off)

See what you get on this one.

- edit - Hey wait a sec, all 107's? That's the color of the Adelson example I posted. Is it possible that's what you're referring to?

thefool · Post by **thefool** » Thu Sep 15, 2005 4:32 pm

i did it with the mask off.

rsts · Post by **rsts** » Thu Sep 15, 2005 4:34 pm

The center sections are extremely pixilated.

While there are pixels that are the same in each one, it's almost impossible to say they're exactly the same, whereas on the first one it was clearly evident they were.

I haven't tried printing them out and "masking" them myself.

thefool · Post by **thefool** » Thu Sep 15, 2005 4:43 pm

exactly. the result cant be 100% equal if you dont have the right spot.
However what i did was:

Open paint.
Paste the screenshot in.
Enable the colour tool.
Select a spot in the first cube. It automatically shifts to the drawing-tool.
Paint over it!
Then you will see, that they are the same colour.

Num3 · Post by **Num3** » Thu Sep 15, 2005 5:53 pm

You can't use Jpegs for testing. Information *is* lost from the original file...

netmaestro · Post by **netmaestro** » Thu Sep 15, 2005 6:21 pm

OK. Time to get PureBasic into this. Here is the third interactive example:

I have carved just the two relevant squares out of the picture and saved them as square1.bmp and square2.bmp. I found them to look identical, but heavily pixellated.

Why would they be made of a mixture of colors? There is no shadowing necessary for these squares, why wouldn't the author just make them all one consistent color? That's what Adelson did in his example, the checkerboard/cylinder.

The only answer I can come up with is that the author of this example tried to push the envelope by slightly altering the colors in a way that favored his desired effect without going so far as to make the squares appear different when viewed interactively.

I wanted to see exactly what was going on, so I wrote this piece of PB code and ran it:

(download prog & images here: http://www.networkmaestro.com/imagetest.zip)

Code: Select all

count.l = 0
total.l = 0
avg.l=0
r1.l=0
g1.l=0
b1.l=0
r2.l=0
g2.l=0
b2.l=0

LoadImage(0,"square1.bmp")
StartDrawing(ImageOutput())
For i = 1 To ImageWidth()
  For j = 1 To ImageHeight()
    If Point(i,j)<>RGB(255,255,255)
      r1 + Red(Point(i,j))
      count + 1
    EndIf
  Next j
Next i
r1=r1/count
count=0
For i = 1 To ImageWidth()
  For j = 1 To ImageHeight()
    If Point(i,j)<>RGB(255,255,255)
      g1 + Green(Point(i,j))
      count + 1
    EndIf
  Next j
Next i
g1=g1/count
count=0
For i = 1 To ImageWidth()
  For j = 1 To ImageHeight()
    If Point(i,j)<>RGB(255,255,255)
      b1 + Blue(Point(i,j))
      count + 1
    EndIf
  Next j
Next i
b1=b1/count
count=0
StopDrawing()

LoadImage(0,"square2.bmp")
StartDrawing(ImageOutput())
For i = 1 To ImageWidth()
  For j = 1 To ImageHeight()
    If Point(i,j)<>RGB(255,255,255)
      r2 + Red(Point(i,j))
      count + 1
    EndIf
  Next j
Next i
r2=r2/count
count=0
For i = 1 To ImageWidth()
  For j = 1 To ImageHeight()
    If Point(i,j)<>RGB(255,255,255)
      g2 + Green(Point(i,j))
      count + 1
    EndIf
  Next j
Next i
g2=g2/count
count=0
For i = 1 To ImageWidth()
  For j = 1 To ImageHeight()
    If Point(i,j)<>RGB(255,255,255)
      b2 + Blue(Point(i,j))
      count + 1
    EndIf
  Next j
Next i
b2=b2/count
count=0
StopDrawing()

Debug "Avg reds in image 1: "+Str(r1)
Debug "Avg reds in image 2: "+Str(r2)
Debug ""
Debug "Avg greens in image 1: "+Str(g1)
Debug "Avg greens in image 2: "+Str(g2)
Debug ""
Debug "Avg blues in image 1: "+Str(b1)
Debug "Avg blues in image 2: "+Str(b2)

The result of this program shows that the overall blues of the square intended to appear yellow are reduced - 138 average blues as opposed to 143 average blues in the other square. The result of this change is too subtle for the naked eye to detect, but it makes for an overall more yellow appearance.

In my humble opinion, the author of these interactive examples had a really good thing to start with - Don't get me wrong, they are excellent illusions. It's just that he obviously tampered with the colors - numbers don't lie - and cast the whole project into question.

MikeB · Post by **MikeB** » Thu Sep 15, 2005 11:46 pm

Looks to me like a classic case of trying to make the facts fit the theory! Running an eyedropper over the squares of the first example shows they are the same, both when done on the screen with an eyedropper utility and after loading a screengrab into Paint Shop Pro. The centres of the cross pieces are harder since there is a load of shading and the positions are not the same but if you make a screengrab and then cut out just the centres and paste them side by side on a black background they certainly look the same which is good enough. You don't need fancy analysis to prove an optical illusion, if one looks yellow and the other looks grey with the background and they look the same without you have demonstrated the effect. Nit picking analysis which shows that there is actually an average shift that the eye can't detect anyway does change the fact that the eye (brain actually) has been fooled into seeing what is not there!

MikeB

netmaestro · Post by **netmaestro** » Fri Sep 16, 2005 12:11 am

What sets the first example apart from the interactive ones is that it's provable and you can win bets with it.

My point about the others is that you can't prove them - not that they don't work, they do. They are excellent. But if you bet someone 20 bucks that they're the same you would lose your money. I just think that the person who authored them should go back and alter them so that they can be shown equal, and I don't think it would detract much from their effectiveness.

Nit picking analysis? Try to take somebody's money when the colors can be shown unequal and they'll smother you in nit picking analysis! Then they'll take YOUR money!

Kale · Post by **Kale** » Fri Sep 16, 2005 12:40 am

netmaestro wrote:What sets the first example apart from the interactive ones is that it's provable and you can win bets with it.

My point about the others is that you can't prove them - not that they don't work, they do. They are excellent. But if you bet someone 20 bucks that they're the same you would lose your money. I just think that the person who authored them should go back and alter them so that they can be shown equal, and I don't think it would detract much from their effectiveness.

Nit picking analysis? Try to take somebody's money when the colors can be shown unequal and they'll smother you in nit picking analysis! Then they'll take YOUR money!

Probably the squares are not the same due to the image compression used. Take a look a jpeg images to see what i mean. Lossy compressions such as jpeg heavily pixelate original smooth images.

MadMax · Post by **MadMax** » Fri Sep 16, 2005 1:12 am

Very good effect, quite clever.

netmaestro · Post by **netmaestro** » Fri Sep 16, 2005 1:22 am

Probably the squares are not the same due to the image compression used.

Sheese Louise that didn't even occur to me. I'll bet you're right. And something like that has probably been knocking around the internet for years appearing in various forms and being copied over and over again.