World Wide Panorama mailing list archive

Mailinglist:wwp@yahoogroups.com
Sender:thinnairstudios
Date/Time:2005-Oct-01 19:31:00
Subject:Re: Image width / 96 and height

Thread:


wwp@yahoogroups.com: Re: Image width / 96 and height thinnairstudios 2005-Oct-01 19:31:00
Hi Dave,

Glad to be of help. Here are a few more things to chomp on. LOL

Angle of view of the Sigma 8mm in vertical orientation is 116 
degrees. Specifically that would be for a 1.5X factor ( 15.6 x 23.7 
CMOS size)

Sigma 8mm
Image width = 2336
Field of view = 116 deg
= 20.1 angular resolution - pixels/degree

For my Nikkor 10.5mm I get 20.2 angular resolution. Very interesting. 
So close.

You state you used the Perspective (12 mm) value. I was looking at 
the Cubic value because we are going to convert to a cubic format in 
the process of making the MOV file.

But if you run some numbers through the calculator the perspective 
12mm value is always the same as the cubic value. Also as you 
increase focal length the linear pixels value decreases. The 12mm 
value is exactly ? of the value for a 24mm perspective lens.

If this is so then that means the cubic format is displayed through 
the viewer at a perspective correction equivalent of a 12mm lens. 
Interesting.

The "panoramic image format" rows for perspective 12mm to 40mm, I 
believe are for panoramic images that will not be displayed in a 
viewer. These images will be outputted as a single image displayed at 
the calculated size.

Now that I think about it that is what all the numbers mean. Except 
that the cubic, cylinder and equirectangular Spherical formats would 
of course be viewed through a viewer. But you could also just display 
them as is if you wanted to.

These formats are confusing me:

Single Round 180? Fisheye
Double Round 180? Fisheye 
Round 360? Fisheye
Anybody know what these are?? I have never heard of a "double round" 
fisheye. And what is the difference between a "round fisheye" and 
a "single round" fisheye?

Normally you would stitch these kinds of images together and get a 
cubic, cylinder and equirectangular image. But I guess if you needed 
to size the image down to make a simple print or just display the 
image on a monitor you would need to know these values to keep the 
angular resolution desired.

Full Frame 180? Diagonal Fisheye

I am not familiar with all the stitching software out there so there 
must me a need to know these numbers for some reason.

Since I use the Nikkor 10.5mm, which is a full frame 180 diagonal 
fisheye I would like to know what this value means. Is there software 
that will correct each frame of a panorama sequence and then stitch 
them together producing a correspondingly wide image? I assume this 
image would just be displayed flat on the screen or printed. Can 
anyone confirm this assumption?


Troy Ward

Dave Wrote:
>>> I suspect that to get a smaller total file size
>>> I might use 1/2 1594 or 797 cube face or TILE SIZE.
>>>
Yes this works. Actually by halving the cube face you are halving the 
angular resolution. In fact it works out the numbers scale equally by 
percentage. For example 10% of 13.9 is 12.51. 12.51 yields a cube 
face of 1435. And if you notice 1435 is exactly 10% smaller than 1594.

With this in mind we now consider the image compression that will be 
used to get to our target byte size. The trick is to set an angular 
resolution that requires none to 10% image compression to get to the 
target byte size. It is my experience that it takes about 20% 
compression of an image before you start to notice any quality 
difference. So keeping things perfect would be no more than 10% 
compression and therefore no more than 10% reduction in angular 
resolution.

We know that jpeg compression is not tied to image size directly. The 
amount of byte savings at a given compression setting depends on the 
image it self. Some will compress a lot while others require twice 
the amount of compression to get to the same byte size. It just 
depends on the tonal range and amount of colors in each image.

It is true that reducing angular resolution simply reduces the size 
of the cube face needed.  But this also reduces the quality.
But consider the intended viewer dimensions

Say we are doing a full screen pano. Max display size on a 1280 x 
1024 monitor is 1280 less browser borders. You could display even 
larger but normally we display at 100% of the browser width for full 
screen. So I am going to use a display size of 1200 wide. We need to 
figure out the "sweet spot" in angular resolution that will make a 
cube face that we can compress 10% and still be below our target byte 
size

Minimum cube face size should be 1000 to 1200. But we want the pano 
to be real nice. We are being picky about the quality. BUT?.we have 
to keep the file size byte size less than 2mb.

We pick a 13.9 angular resolution which equals a 1594 x 1594 cube face
Now, say this image is 2.8 mb when outputted to a MOV file.
We can reduce the angular resolution; increase the compression or 
both to get below 2mb.

I think it is better to keep as much angular resolution as possible 
and go with a little more compression when dealing with full screen 
images.

Markus wrote:
"- the larger the image is, the less noticable the JPG
compression "residue" will be (usual JPEG blocks are 8x8),
so i may get away with choosing less JPG quality"

With this in mind when dealing with smaller display images one should 
choose a smaller angular resolution and less compression to control 
byte size.



--- In #removed#, "Dave 360texas.com" <#removed#> wrote:
> Troy -  I think I understand what you are saying.
>   
> I use a Canon 20d with a Sigma 8 and a portrait orientation.
> Image size of 3504h x 2336w and 4 images/360?
> Equirectangular stitched image 5000 x 2500 2:1
> Angular Resolution 13.9 pixels/degree
> Panorama Image Format Perspective (12mm) 1594 pixels cube face 
> 1594x1594=2,540,836 total pixels cube face
> 2,540,836 x 6 cube faces = 15,245,016 total pixels
> So for converting it to a QTVR to achieve a 1:1 I should use 1594.
> 
> I suspect that to get a smaller total file size 
> I might use 1/2 1594 or 797 cube face or TILE SIZE.
> 
> Believe it or not... this makes sense. Thank you for taking the 
time 
> walk through the exercise.
> 
> Regarding your comment about the proper viewer dimensions.  I think 
> the viewer should be about 30% of the image height.  2500h x 30% = 
> 750 pixels high.  What viewer height do find provides best 
rendering?
> 




Next thread:

Previous thread:

back to search page