Ad
Ad
Ad
Pages: [1]   Bottom of Page
Print
Author Topic: Resolution When Transforming a Tiff into a Sphere  (Read 758 times)
deanwork
Sr. Member
****
Offline Offline

Posts: 722


« on: January 06, 2012, 03:16:35 PM »
ReplyReply

I'm trying to transform three grayscale files from drum scans into large spheres. I would like to make 40x40 prints of these or at least 30x30. This is a show of large prints of geological formations shot on 100 iso 4x5 drum scanned film for large bw prints. It is about plate tectonics.

I have been successful in obtaining what I want either using Photoshop CS 5 and using the 3D sphere filter, or by doing this by hand the old way with using a free transform and shaping it to a sphere myself with handles.

But in both cases I'm loosing a lot of resolution and with extra sharpening it isn't there yet. Actually the old way is giving me a little bit better resolution it seems but of course takes a whole lot longer.

Does anyone know of somthing I can do to improve the sharpness when creataing these kinds of files. They are all super high quality drum scans from 4x5 Tmax 100 scanned at 4000 ppi on an Aztek Premier. I see people are doing 3D sphere panoramas, apparently successfully.

john

« Last Edit: January 06, 2012, 07:00:14 PM by deanwork » Logged
Guillermo Luijk
Sr. Member
****
Offline Offline

Posts: 1291



WWW
« Reply #1 on: January 06, 2012, 03:29:30 PM »
ReplyReply

Some examples of the resolution loss you are experiencing would help.

I can tell you that some time ago I did some exercises in transforming an image into some crazy shapes (spheres, waves, spiral) and found that upsampling previously the image in Photoshop to a size N times larger than the ouput size, allowed me to do these transformations using a hyper-basic nearest neighbour transformation (the only one my programming skills can afford), and still get a very high quality output.

Some examples here. As expected, straight nearest neighbour produced aliasing and information loss, but resampling the images prior to the basic transformation radically improved the result.

left: input image, centre: straight nearest neigbour transform, right: upsizing + nearest neighbour transform + downsizing.

SPHERE MAPPING


WAVE MAPPING


SPIRAL MAPPING



Maybe if you previously upsize your scans, then apply the transform, and finally downsize to the final format can improve your results.

Result
« Last Edit: January 06, 2012, 03:33:44 PM by Guillermo Luijk » Logged

deanwork
Sr. Member
****
Offline Offline

Posts: 722


« Reply #2 on: January 06, 2012, 06:53:46 PM »
ReplyReply

Thanks Guillermo,

I tried to resample image before transformation and then downsize and it didn't change anything here.

Essentially I'm starting with aprox 500 meg 16 bit grayscale files. After creating a sphere with the 3D PS mechanism you end up with 3 times that big of a file, but when it is cropped down to the circle the size stays pretty much the same. In all cases with many files it is loosing definition a lot and becomes blurry. When I downsize it in half it is still unacceptable. When I downside it in half again it is decent but still not where the original file was in sharpness. In other words a drum scan that I'm using for a 40x60 print is only acceptable when converted this way to do say a 10x10 print.

The only way I can see doing this for the three files I want to do this way is to scan them at 8000 ppi and make a 20x20 print, if that works even.

Apparently when transforming a file (which is all about texture in this case) to a sphere in CS5 using 3D transformation the compression and or stretching of pixels does a lot of damage to the file. I've noticed some of this when doing ordinary stretching of a size in image size to elongate it, but no where near this degree. Like I said before I'm getting quite a bit better results doing a circular transformation by hand using the free form transform method in PS, but it is far from great in its impact on sharpness also, but better than this.

Here are a couple of before and after examples. Now that they are reduced down to a jpeg you can't tell how poor the sharpness was.

The show is about the geology of plate tectonics, and I only want to do 3 of the 25 prints this way really, but I'd like to do them as large as the others if possible.

john

« Last Edit: January 06, 2012, 07:03:44 PM by deanwork » Logged
aduke
Sr. Member
****
Offline Offline

Posts: 365


« Reply #3 on: January 06, 2012, 10:59:07 PM »
ReplyReply

I wonder if this is just the difference in logical area. The area of the square image is about 1/3 of the area of half a sphere, i.e., the area of the square with side D is D**2, the area of half a sphere is
pi * D**2, where pi = 3.14159... .


Alan
Logged
Pages: [1]   Top of Page
Print
Jump to:  

Ad
Ad
Ad