|
|
Post by Peterj on Sept 3, 2018 18:00:50 GMT
In this tutorial Ian Norman shows us how to enhance the resolution of a camera sensor with a technique called superresolution. With this technique, it’s possible to mimic the sensor-shift high-resolution mode found on cameras like the Olympus OM-D E-M5 Mark II to squeeze more megapixels out of the camera sensor. In his example, he increases the resolution of a 24 megapixel photo to more than 90 megapixels. See the full write-up and video walkthrough in this tutorial.
Using PSE one might be able to duplicate this procedure; Elements+ will make this simple using the Noise Stacking script.
|
|
|
|
Post by hmca on Sept 3, 2018 18:33:33 GMT
Look forward to watching this when time allows.....sounds very interesting!
|
|
|
|
Post by kdcintx on Sept 3, 2018 21:28:14 GMT
Thanks Peter. I want to try this technique. You sure do find some interesting stuff to keep us learning.
|
|
|
|
Post by Sydney on Sept 4, 2018 0:03:59 GMT
Thanks for sharing Pete. You are a real super sleuth!
|
|
|
|
Post by Bailey on Sept 4, 2018 0:57:43 GMT
Hi Peter, I have read through the tutorial - interesting. But as is mentioned in the tutorial:
But it does have other benefits, but they also can be achieved by other means.
In any case, after evaluating AI Gigapixel, for me AI Gigapixel involves much less processing set up and can output much higher pixel count images than SuperResolution.
Also, the only time I would ever need/consider enlarging an image is for printing purposes. Both SuperResolution and AI Gigapixel are imho gross over-kill for enlarging/enhancing images to be displayed on screens as I discuss in my Sizing Images For Web Display thread. For example, my screen is 1920px wide. So any online image that is more than 1920px wide is going to be downsized by my browser (resulting in pixels being discarded) to whatever space has been allocated to the image by a web page's HTML and CSS.
Reading through the technique in the tutorial, my gut feeling is that the resultant clarity/sharpness of the output image might not be as high quality as AI Gigapixel. But if I can make some time in the future I will try to do a comparison out of interest.
|
|
|
|
Post by Peterj on Sept 4, 2018 1:18:45 GMT
Bailey I appreciate your advanced photographic knowledge. This thread has absolutely nothing to do with ... In any case, after evaluating AI Gigapixel, for me AI Gigapixel involves much less processing set up and can output much higher pixel count images than SuperResolution.
You seem to always have alternate ideas and present them in other folks' threads.
I really don't appreciate your comments in this thread - to me it's just a method to derail the subject to an area for you to expound on something you feel strongly about.
Please start you own thread to
and keep your comments focused on sensor shift and super resolution - otherwise stay out.
|
|
|
|
Post by Bailey on Sept 4, 2018 1:30:25 GMT
Hi again Peter, No problem  My purpose was to simply keep this technique in perspective, especially for newbies to photography and/or post processing, relative to other options that can achieve the same and better output results.
The article you posted was originally posted in Feb. 2015, so it's hardly ground breaking new technology or a new technique. Imo, it's been well and truly superseded.
As the tutorial also states:
If my post is way off-topic, I am sure one of the administrators or moderators will delete it. I'll leave you to it now |
|
|
|
Post by Peterj on Sept 4, 2018 5:10:07 GMT
I decided to try this procedure with these testing parameters: 1) shoot at camera sweet spot - base ISO =100, f/4, AWB aperture priority 2) shoot hand held - at least 20 images in burst mode [didn't try to be extremely steady] 3) raw images opened in ACR - no adjustments [tut suggests upscaling - not done for 1st test] 3a) 2nd test -original raw files batch processed in PSE to 300% upscale using nearest neighbor down side here is output psd fies to load into noise stacking 4) opened all 23 images and executed Elements+ noise stacking script 5) use PSE18 to adjust view for comparison 6) screen shots of comparison included
I see benefit in the no upscaling and if you really need a huge file 300% upscaling provides more detail; I'm not certain I'll be using the upscaling.
The method in the original tutorial seemed to be a bit more complicated than the method I used in PSE. Perhaps someone can try the PS method ... if anyone wants to use my original files let me know and I'll provide.
 |
|
|
|
Post by Bailey on Sept 4, 2018 10:59:14 GMT
On closer reading of the tute I see there is an error/inconsistency in what they expect from this technique. For the sake of accuracy and eliminating confusion especially for any newbies following this thread or for anyone where maths is not their strong suite  , in the text of the tutorial - 200% should be 100% because doubling the length and width to get 4x the pixel count is a 100% increase on each of the length and width and not 200%. So if you have a 500px x 500px image (250,000px) you probably won't get an image larger than 1000px x 1000px (1,000,000px) with this technique according to the tutorial.
|
|
|
|
Post by Peterj on Sept 4, 2018 15:29:29 GMT
Posting some more details about the methods and results posted earlier.
PSE18 was used to upscale and view [percentages were calculated by PSE]
original file size 14.3 MB - noise stacked file size 69.2 MB
- 300% upscaled file size 388.2 MB
|
|
idea for increasing your camera's resolution using hand held burst shots and PS.
idea for increasing your camera's resolution using hand held burst shots and PS.
, in the text of the tutorial -