It's the photographer not the camera

rutt

We always say this, but I want to share one of the best examples I've ever experienced. In the mid 90s I worked at Silicon Graphics and there was this guy there, Paul Haeberli, who was a real bonafide genius, both as a computer scientist and as an artist. He just had that artist's eye and creativity that made me wonder why I even try.

Pretty early in the history of digital cameras, maybe 1995, he asked around. He wanted a digital camera, but he didn't want to spend very much, maybe $500. (Silicon Graphics would have bought Paul anything at the time, all it had to do was exist.) I don't know what he got, but it couldn't have had much more resolution than a cell phone camera, and I bet it took an eternity between the time he pushed the shutter and it actually captured the image. But he instantly went out and took at least 100 pictures I was jealous of. I dug up a few of these on the web:

http://www.sgi.com/misc/grafica/gallery/index.html

How about this for a self portrait? Think it could make the finals?

http://www.sgi.com/misc/grafica/gallery/index.html

A lot of Paul's stuff, not just photographs, but also playful projects and even pretty deep computer graphics ideas are collected here:

http://www.sgi.com/misc/grafica/

If you look closely enough, you can figure out that Paul really invented a lot of stuff about computer graphics which we take for granted now.

gluwater

That is really cool rutt. Thanks for letting us see it. I really like the "Paper and Light" shot. Creativity is only limited by our imagination, not our equipment.

Andy

he's done fine work - thanks for pointing it out, rutt!

David_S85

One really smart dude
His photography is interesting. I looked around there for a while, and found this...

A Multifocus Method for Controlling Depth of Field
http://www.sgi.com/misc/grafica/depth/index.html

Could someone way smarter than me explain how I can accomplish that using layers (I'm using PSP) without buying any software?

Andy

David_S85 wrote:

Could someone way smarter than me explain how I can accomplish that using layers (I'm using PSP) without buying any software?

look at lordvet's posts:

http://www.digitalgrin.com/showpost.php?p=156830

http://www.dgrin.com/showthread.php?t=18345

David_S85

Yup. Thanks Andy. I read both of those posts before. I'm just trying to go the long way about it -- through the manual steps to try to learn the process that the software takes to do this. I know. Stupid, huh?

andy wrote:

look at lordvet's posts:

http://www.digitalgrin.com/showpost.php?p=156830

http://www.dgrin.com/showthread.php?t=18345

XO-Studios

David_S85 wrote:

His photography is interesting. I looked around there for a while, and found this...

A Multifocus Method for Controlling Depth of Field
http://www.sgi.com/misc/grafica/depth/index.html

Could someone way smarter than me explain how I can accomplish that using layers (I'm using PSP) without buying any software?

OK not being a slouch in PS myself, I can follow that technique right up to the point of:

Now we compare the two edge images, and make an image that is black where the left image has more edge information, and is white where the right image has more edge information.

I even understand the step after that, but have no clue how to establish this one. I'd be more likely to use masks in each image, but this seems to work better. So by all means if anyone has any thoughts/ideas how to establish that step, I'd be very interested in hearing that. (Other than packaged software)

XO,

rutt

XO-Studios wrote:

OK not being a slouch in PS myself, I can follow that technique right up to the point of:

Now we compare the two edge images, and make an image that is black where the left image has more edge information, and is white where the right image has more edge information.

I even understand the step after that, but have no clue how to establish this one. I'd be more likely to use masks in each image, but this seems to work better. So by all means if anyone has any thoughts/ideas how to establish that step, I'd be very interested in hearing that. (Other than packaged software)

XO,

I don't know how Paul actuially did this. He didn't necessarily use Photoshop. He had his own graphics toolkit for SGI that he wrote himself and also he could use GL to get SGI graphics hardware to do a lot. You need a lot of horsepower for rendering, more even than PS gives you.

But I think it might be possible to do this in PS. I'll play with it a bit.

rutt

XO-Studios wrote:

OK not being a slouch in PS myself, I can follow that technique right up to the point of:

Now we compare the two edge images, and make an image that is black where the left image has more edge information, and is white where the right image has more edge information.

I even understand the step after that, but have no clue how to establish this one. I'd be more likely to use masks in each image, but this seems to work better. So by all means if anyone has any thoughts/ideas how to establish that step, I'd be very interested in hearing that. (Other than packaged software)

XO,

I couldn't figure out how to reproduce Paul's exact series of steps in PS, but if you are willing to look at the bigger picture, it's pretty easy to use the germ of his idea to accomplish the same thing.

After the subtractions, Paul has these two images:



I just used the image on the left as a layer mask for the leftmost orignal (the one with closer focus) and the image on the right as a layer mask for the rightmost original (the one with farther focus). So I had 2 layers, each showing only the in focus parts of each image. Then I just needed either original image as a bottom layer to show the parts without any particular edges.

As I said before, I doubt Paul was working with Photoshop. And I think he was thinking in the context of SGI's graphics primitives. If the solution could be formulated in terms of GL primitives, then it could performed in real time, perhaps with two video camera feeds as input.

XO-Studios

OK here is where I am at so far.

One challenge is the original images he uses are in GIF, that doesn't help.

Anyway take each image, in a seperate layer
Call one far focus, the other one close focus.
Now duplicate each layer, take the top layer, apply gaussian blur to taste and set blend to difference. Voila you now have the blur subtracted from the original.

Now how do I get from here to the nice white/black mask.

Looking at the articles date (1994) I am pretty sure that PS today on a regular lap/desktop far exceeds the SGI power used in the original. I know I am close, as soon as I can create the pixelated image, I useone as is as a mask and the other inverted as a mask

BTW, Smart Blur edge only will give similar results.

XO,

rutt

XO-Studios wrote:

Looking at the articles date (1994) I am pretty sure that PS today on a regular lap/desktop far exceeds the SGI power used in the original.

It's not a question of computer horsepower, it's the fact that Haeberli was a graphics programmer using a different set of primitives than Photoshop's and also free to combine them in more complex ways (because he was actually programming instead of pointing and clicking.)

Anyway, didn't the layer mask idea make sense to you? Or do you still want to solve the puzzle of how to put the pieces together exactly the same way and in the same order as Paul did?

XO-Studios

rutt wrote:

It's not a question of computer horsepower, it's the fact that Haeberli was a graphics programmer using a different set of primitives than Photoshop's and also free to combine them in more complex ways (because he was actually programming instead of pointing and clicking.)

Anyway, didn't the layer mask idea make sense to you? Or do you still want to solve the puzzle of how to put the pieces together exactly the same way and in the same order as Paul did?

Your explanation made perfect sense, but you hit the nail on the head, putting the pieces together it was done originally is somewhat of a challenge, also as it seems to be a better technique to build upon. There are various things that this technique when properly applied and learned would do miracles on. One application which I already foresee with what I have now, is a mask that can be used on softening portraits while conserving important details.

Forever the engineer.

XO,

rutt

XO-Studios wrote:

Forever the engineer.

That makes two of us. I'll have to think about whether the layer mask approach is really any less powerful than Paul's original approach. In fact, I came up with a simplification: just use one mask to overlay the in-focus parts of one image on the complete other image.

Perhaps the layer mask approach is actually more powerful. Maybe we could try to play with some images in a separate thread on the PS forum?

I was thinking about the comparison of early 90s SGI hardware vs current hardware. The comparison you wanted to make isn't quite right. I'm sure that if you compared today's $100 graphics card to the graphics engine in SGI's then $100k machines, today would win. But if you compare PS today to SGI graphics engine then, it's not quite right because PS today doesn't use the graphics hardware, just the CPU. So it doesn't get the advantages of massive parallelism and specialized operations which the old SGI boxes had.
Not that it matters. Forever the engineer.

rutt

Forever the engineer, I just couldn't leave this alone. We are wondering about this step in Paul's description:

Paul Haeberli wrote:

Now we compare the two edge images, and make an image that is black where the left image has more edge information, and is white where the right image has more edge information.

Paul's description of the blending of the two above images isn't strictly accurate. For example, consider the area between the far bottle and the cog wheel. That is black in both images. But in the blend, it is speckled. More importantly, consider the label on the near bottle. The black area is not just the part that is white on teh leftmost input image. It seems there was some sort of blur and then threshold adjustment. My conclusion is that Paul hasn't really described exactly what he did very wel.

It doesn't really matter. Suppose we take only one of the input images and use it to mask it's corresponing original and layer it on top of the other original. Now we'll get the sharp edges from the upper layer and the rest of the lower image.

devbobo

rutt wrote:

Forever the engineer, I just couldn't leave this alone. We are wondering about this step in Paul's description:



Paul's description of the blending of the two above images isn't strictly accurate. For example, consider the area between the far bottle and the cog wheel. That is black in both images. But in the blend, it is speckled. More importantly, consider the label on the near bottle. The black area is not just the part that is white on teh leftmost input image. It seems there was some sort of blur and then threshold adjustment. My conclusion is that Paul hasn't really described exactly what he did very wel.

It doesn't really matter. Suppose we take only one of the input images and use it to mask it's corresponing original and layer it on top of the other original. Now we'll get the sharp edges from the upper layer and the rest of the lower image.

To me it looks like this last step could have been produce using something like the Threshold tool is this correct ?

rutt

devbobo wrote:

To me it looks like this last step could have been produce using something like the Threshold tool is this correct ?

I don't thnk so since the background is uniformly black in both input images. Threshold would keep it that way.

USAIR

Wow thanks Rutt for the great link
This guys work is amazing
Love the composites very nice

They say that imitation is the sincerest form of flattery
I hope so because this guy has given me lots of ideas

Fred