Stair Interpolation resize methods compared..

[Fulvio Senore]

Fulvio,
Thanx for the link. I though I had read about everything there was
but this is good. I wish PS had a way to make a custom kernel for
its bicubic interpolation routine. Something along the lines of the
custom convolution kernel.
VG

I have done some tests (as you know) and I have found that PS bicubic interpolation is probably the best routine that could be written using that support (number of points used to compute each pixel). All other programs are, at different extents, worst performers. For this reason i don't believe that allowing the user to use a custom kernel could lead to better results.

To have better results you must use a wider support, and it doesn't work with every kind of image. PS recubic is a good compromise that works well with any kind of image, not only photographs.

--
Fulvio Senore

 

[VG]

Fulvio,

You are dead on it. What I was wishing for was a kernel that allowed manipulation of the pixel weighting as well as the depth of pixels that PS would wade into for interpolation data.
Thanx,
VG

Fulvio,
Thanx for the link. I though I had read about everything there was
but this is good. I wish PS had a way to make a custom kernel for
its bicubic interpolation routine. Something along the lines of the
custom convolution kernel.
VG

I have done some tests (as you know) and I have found that PS
bicubic interpolation is probably the best routine that could be
written using that support (number of points used to compute each
pixel). All other programs are, at different extents, worst
performers. For this reason i don't believe that allowing the user
to use a custom kernel could lead to better results.

To have better results you must use a wider support, and it doesn't
work with every kind of image. PS recubic is a good compromise that
works well with any kind of image, not only photographs.

--
Fulvio Senore

 

[VG]

Well, I've pretty much finished the first round of the SI testing using a portrait based image. I'm compiling my data (I thought I gave up programming) and hope to post some more of the results. I've found some interesting things but the gist of it is:

There aren't big differences between percentages and where there is, it is skewed toward the useless instead of useful.
The 10% touted around ain't the magic number.
There is no magic number.
Percentages to 5 decimal places seem to work the same as whole numbers.
Regardless of what you see on the screen, it's gonna look different in print.
All of the output is totally subjective.
You can draw whatever conclusions you like.
If I see one more eyeball at 400%....

Regards,
VG

 

[ZMR]

VG,

If you "compile" all you results into a stack of layers and set every layer except for the background in DIFFERENCE mode, and then compile a stack of the difference histograms, are there counts any beyond level 20? Of course, I mean you compare only two at a time.

 

[VG]

There aren't enough differences to attach any significance to. It's mainly a haloing issue is all.
VG

VG,

If you "compile" all you results into a stack of layers and set
every layer except for the background in DIFFERENCE mode, and then
compile a stack of the difference histograms, are there counts any
beyond level 20? Of course, I mean you compare only two at a time.

 

[WODIN]

Your findings are distressing. So all those SI tiny tweaks and specific amounts of % modifications make little or no difference. What does then? Till you came along many of us sought comfort in ignorance. Wouldn't it have been better to allow all us 'little people' alone?

signed,
'sad-sack - me-sha-da-ben-digo'
--
So you ain't got all ya want. Doncha know
a dream's the stuff happiness is made of.
Wodan
Canon Pro 70
http://www.pbase.com/image/8248718
http://www.pbase.com/image/4072188
http://www.pbase.com/image/11604500

 

[ZMR]

Your findings are distressing. So all those SI tiny tweaks and
specific amounts of % modifications make little or no difference.
What does then? Till you came along many of us sought comfort in
ignorance. Wouldn't it have been better to allow all us 'little
people' alone?

I still believe that there is an optimum number of steps for a given enlargement factor. Just think about it: too many steps will defintely cause too much image degradation, while too few steps will be more like an one-step fast job. So there must be a point that's a compromise of the two. The curve could be very flat, however.

 

[VG]

NO, I didn't say that all tweaks make no difference. I did say that it is subjective though. There are just so many factors dealing with a scaling algorithm. Image size, detail, noise, ISO, color, tonality all effect what you will perceive as ideal.

This was just the first round and it laid some groundwork for future meddlings. It is a good idea to have your eyes opened once in a while and not just blindly follow the crowd. I know that this session has opened mine a lot wider.
VG

Your findings are distressing. So all those SI tiny tweaks and
specific amounts of % modifications make little or no difference.
What does then? Till you came along many of us sought comfort in
ignorance. Wouldn't it have been better to allow all us 'little
people' alone?

signed,
'sad-sack - me-sha-da-ben-digo'
--
So you ain't got all ya want. Doncha know
a dream's the stuff happiness is made of.
Wodan
Canon Pro 70
http://www.pbase.com/image/8248718
http://www.pbase.com/image/4072188
http://www.pbase.com/image/11604500

 

[Bill Richardson]

It is better to make your own. You need several, and they have to
allow you to end up with a target size from a given source.
Everyone’s needs are different and the actions are really easy to
make.

I have settled on 5% increments. I start an action named 5 X 5 and
resize an image 5 times at 5%. I start another action named 10 X 5
and run the first action 2 times. I make another called 20 X 5 and
run the second action twice. Those three actions usually cover my
normal resizing. I can run them in tandem or just combine them to
make a new action if my needs change.

Just make sure to use the % in the resize so it will work on any
image.

Someone suggested throwing in a little UM with SI. I might try a
little after the first 5 so that I apply some UM each 5 steps.

--slipe, Good idea. However, I crop a bunch of photos to different ppi end results and want them all to come out as 4x6", 300ppi. Is there any way to create an action to target this end point with different starting ppi's?
Bill Richardson
Barrington, IL (USA)
Canon G3, 420EX, and PhotoShop 7

 

[Bill Richardson]

Regardless of what you see on the screen, it's gonna look different
in print.
Regards,
VG

That is so true and frustrating!
Bill Richardson
Barrington, IL (USA)
Canon G3, 420EX, and PhotoShop 7

 

[aruzinsky]

I was walking down the street at night and came upon a big crowd of people looking for something under a street light. I asked them, "What are you looking for?" They all replied, "Our wallets." I said, "You all lost your wallets under this street light?" They replied, "No, we each lost our wallet in a different part of the city." I asked, "Then why are you all looking here?" They replied in unison, "Because the light is better here."
--
Author of SAR Image Processor
http://www.general-cathexis.com

 

[VG]

Hehehe, aruzinsky, that's a good one. SAR Image Processor is in line to compare to the other methods, that is if it can handle the file sizes we're talking about.
VG

I was walking down the street at night and came upon a big crowd of
people looking for something under a street light. I asked them,
"What are you looking for?" They all replied, "Our wallets." I
said, "You all lost your wallets under this street light?" They
replied, "No, we each lost our wallet in a different part of the
city." I asked, "Then why are you all looking here?" They replied
in unison, "Because the light is better here."
--
Author of SAR Image Processor
http://www.general-cathexis.com

 

[Joe-TN]

Well, for instance, a 35mm slide scanned at 4000ppi would do it. 35mm scanned on the Epson 3200 at 3200ppi produces a file of around 38MB if my experience is typical.

Joe

I'm curious as to what source would provide you with a 40 meg fille
(referance to Genuine Fractals).

--
Joe-TN

http://www.pbase.com/joe_tn/shared_photos

 

[Vendra]

I was walking down the street at night and came upon a big crowd of
people looking for something under a street light. I asked them,
"What are you looking for?" They all replied, "Our wallets." I
said, "You all lost your wallets under this street light?" They
replied, "No, we each lost our wallet in a different part of the
city." I asked, "Then why are you all looking here?" They replied
in unison, "Because the light is better here."

Then a guy across the street yelled to me and questioned, "What are you bothering those busy people for?" I said, "I was curious about what they were doing." He said, "Was it a large croud that caught your attention?" I replied, "Not really, I just felt compelled to find out what was going on." He asked "Are you satisfied?" I answered "Only when they go away so I can look for my wallet."

Vendra