Scottes
16th of July 2004 (Fri), 20:45
I took two photos and upsized them in a variety of ways:
1 Step Bicubic Sharper
1 Step Extensis Smart Scale
11 Steps, 10% upsize per step, using Bicubic
11 Steps, 10% upsize per step, using Bicubic Sharper
11 Steps, 10% upsize per step, using Bicubic Smoother
11 Steps using my Smoother, Smoother, Sharper
That is, 11 steps of 10% upsize per step, using Bicubic Smoother followed by Bicubic Smoother followed by Bicubic Sharper. This 3-step process was done 3 times, followed by a Bicubic Smoother and finishing with a Bicubic Sharper
The process yielded a 285% upsize - a 200x200 going to 570x570, and a 300x300 going to 855x855
When done I layed out the crops onto a single image and printed this on my Epson 1280 at 360 DPI using Epson Premium Glossy Photo paper - 1 4x6 and a 5x7 since the second test used a larger crop.
When they were done, I picked them apart, then showed the prints to my wife. Neither of us knew which image was which method until we had both scrutinized the prints. (It didn't take me long to completely forget the order they were in...) We both agreed exactly on the best 2 and worst 2, but disagreed both times on the 3rd and 4th best. But who really cares about the 3rd and 4th best anyway?
I'm not posting the final upsized images because they're too big, it's too much work, and it's really the prints that prove it. So you'll just have to believe my wife & I.
The following two images were done. Well, actually just the crops shown below were done. The process is a memory and time pig, so there was no way that I was going to do these on 3000x2000 images. These are 100% crops that have had the normal image processing and some sharpening.
http://www.itsanadventure.com/postimages/0425_Org_Cropped.jpg
http://www.itsanadventure.com/postimages/4295_Org_Cropped.jpg
I wanted to try two completely different images: One with a lot of sharp detail - the side of the house and porch mixed with the smoother shapes of the cars. The other was a highly detailed head of a sparrow.
The print shop where I've been taking stuff loved the sparrow because it was so sharp and they printed it the other day at 12x18 @ 400 DPI. I had used my 3-step Smoother Smoother Sharper technique for that print, and I was very happy with the results. I also think it's a good shot for this test - at least for me since I take a lot of shots of furred & feathered wildlife.
So, the results of this test using 6 different processes....
For the house, the #1 choice was the 1-step Bicubic. The second choice - which was very very close - was Extensis SmartScale. Both 11-step Sharper and 11-step Smoother sucked. Who really cares about #3 and #4?
For the sparrow, the #1 choice was my 3-step process, followed by 1-step Bicubic. Both 11-step Sharper and 11-step Smoother sucked. Who really cares about #3 and #4?
In all cases it was *very* difficult to choose the best when the print was held at arm's length.
So what does all this prove? I think the only thing it really proves is that every image is different.
However, I know you want more meat than that, so...
1) In my opinion my 3-step Smoother Smoother Sharper is the best for stuff like fur and feathers, or if noise is a concern. However, I would not use it on images with sharp detail lines like houses and other man-made objects.
2) Smartscale is very good at sharply defined objects with clear color transitions, but overall I don't think it's worth the money. From what I've seen I'll include Genuine Fractals to be in the same category. Even though I haven't actually tested it, I've seen several tests by others.
3) If you're lazy, not so finicky, pressed for time, or don't have enough RAM, then 1-Step Bicubic is plenty good enough for anything that will be more than a foot away. And I'm saying that after upsizing by 285% - which is pretty crazy unless you have a *super* sharp image to start. If you stick to anything less than 200% then it will be very very good.
4) Both 11-step Sharper and 11-step Smoother suck.
5) If you have a noisy image, then SmartScale and my 3-Step do the best job of retaining detail without blowing out the noise. The type of image detail will decide which one will be better.
I did play with my idea for upsizing using multi-step 10% increases using Bicubic Smoother with each step followed by controlled sharpening. I tried controlling sharpening by using an inverted blurred edge mask so that midtones would be mildly sharpened after each 10% increase, leaving edges to be *very* mildly sharpened after each 10% increase.
The idea is simply too complex to be of any use. I won't get into it. But 4 trials runs on a single image was a lot of effort for very little gain. The final result was good but would require *much* testing in order to develop any kind of intelligent algorithm that could handle it. There's too many variables, really.
1 Step Bicubic Sharper
1 Step Extensis Smart Scale
11 Steps, 10% upsize per step, using Bicubic
11 Steps, 10% upsize per step, using Bicubic Sharper
11 Steps, 10% upsize per step, using Bicubic Smoother
11 Steps using my Smoother, Smoother, Sharper
That is, 11 steps of 10% upsize per step, using Bicubic Smoother followed by Bicubic Smoother followed by Bicubic Sharper. This 3-step process was done 3 times, followed by a Bicubic Smoother and finishing with a Bicubic Sharper
The process yielded a 285% upsize - a 200x200 going to 570x570, and a 300x300 going to 855x855
When done I layed out the crops onto a single image and printed this on my Epson 1280 at 360 DPI using Epson Premium Glossy Photo paper - 1 4x6 and a 5x7 since the second test used a larger crop.
When they were done, I picked them apart, then showed the prints to my wife. Neither of us knew which image was which method until we had both scrutinized the prints. (It didn't take me long to completely forget the order they were in...) We both agreed exactly on the best 2 and worst 2, but disagreed both times on the 3rd and 4th best. But who really cares about the 3rd and 4th best anyway?
I'm not posting the final upsized images because they're too big, it's too much work, and it's really the prints that prove it. So you'll just have to believe my wife & I.
The following two images were done. Well, actually just the crops shown below were done. The process is a memory and time pig, so there was no way that I was going to do these on 3000x2000 images. These are 100% crops that have had the normal image processing and some sharpening.
http://www.itsanadventure.com/postimages/0425_Org_Cropped.jpg
http://www.itsanadventure.com/postimages/4295_Org_Cropped.jpg
I wanted to try two completely different images: One with a lot of sharp detail - the side of the house and porch mixed with the smoother shapes of the cars. The other was a highly detailed head of a sparrow.
The print shop where I've been taking stuff loved the sparrow because it was so sharp and they printed it the other day at 12x18 @ 400 DPI. I had used my 3-step Smoother Smoother Sharper technique for that print, and I was very happy with the results. I also think it's a good shot for this test - at least for me since I take a lot of shots of furred & feathered wildlife.
So, the results of this test using 6 different processes....
For the house, the #1 choice was the 1-step Bicubic. The second choice - which was very very close - was Extensis SmartScale. Both 11-step Sharper and 11-step Smoother sucked. Who really cares about #3 and #4?
For the sparrow, the #1 choice was my 3-step process, followed by 1-step Bicubic. Both 11-step Sharper and 11-step Smoother sucked. Who really cares about #3 and #4?
In all cases it was *very* difficult to choose the best when the print was held at arm's length.
So what does all this prove? I think the only thing it really proves is that every image is different.
However, I know you want more meat than that, so...
1) In my opinion my 3-step Smoother Smoother Sharper is the best for stuff like fur and feathers, or if noise is a concern. However, I would not use it on images with sharp detail lines like houses and other man-made objects.
2) Smartscale is very good at sharply defined objects with clear color transitions, but overall I don't think it's worth the money. From what I've seen I'll include Genuine Fractals to be in the same category. Even though I haven't actually tested it, I've seen several tests by others.
3) If you're lazy, not so finicky, pressed for time, or don't have enough RAM, then 1-Step Bicubic is plenty good enough for anything that will be more than a foot away. And I'm saying that after upsizing by 285% - which is pretty crazy unless you have a *super* sharp image to start. If you stick to anything less than 200% then it will be very very good.
4) Both 11-step Sharper and 11-step Smoother suck.
5) If you have a noisy image, then SmartScale and my 3-Step do the best job of retaining detail without blowing out the noise. The type of image detail will decide which one will be better.
I did play with my idea for upsizing using multi-step 10% increases using Bicubic Smoother with each step followed by controlled sharpening. I tried controlling sharpening by using an inverted blurred edge mask so that midtones would be mildly sharpened after each 10% increase, leaving edges to be *very* mildly sharpened after each 10% increase.
The idea is simply too complex to be of any use. I won't get into it. But 4 trials runs on a single image was a lot of effort for very little gain. The final result was good but would require *much* testing in order to develop any kind of intelligent algorithm that could handle it. There's too many variables, really.