Adobe Super Enhance
Posted: Thu 02 Sep 2021, 09:46
When Adobe introduced its "Super Enhance" feature a while back, I did not pay too much attention to it. My main interest was in seeing whether it could help with some images taken on a 6MP Canon EOS 300D where I wanted to crop into the image beyond the sensor's capabilities to resolve detail. The results did not much impress, so I moved on. Unless printing very large or cropping heavily, cameras have long been providing more resolution than most people need. I accept that is a controversial statement for some but the reality is that in most instances the final image is being downsampled when we output it. Images posted to the web are usually low resolution and a 1600 x 1200 DPI is less than 2 MP. For typical camera club use, the standard mount size of 50cm x 40 cm means that when printed most images do not fill an A3 sheet to allow space for the mount border. 10-12 MP is more than sufficient for the purpose, even with a judge examining a print more closely than would happen normally. Resolution of detail when viewed on a screen at 100% does not translate into an improvement to image quality.
Nevertheless, I was intrigued by an analysis of DPReview about the Super Enhance feature, despite the clickbait title. I took a few things away from it. There might be an improvement to image quality by upscaling with Super Enhance then downsampling it afterwards. In the comments to the article, Barney Britton made the point that images made using cameras with Bayer sensors are already to subject to interpolation. In other words, processing commences with the development software making a best guess about the reality of what was captured. Add in camera or profile adjustments for distortion or vignetting which lens designers often employ and the "purity" principle to which some adhere begins to look hopelessly idealistic. All that really matters is what the result looks like and Super Enhance is just another tool to help reach that goal. No one should have any concerns about how it was achieved.
The article also states that Adobe has recruited Marc Levoy, an expert in computational photography, from Google. The intention is clear. Super Enhance represents a trend where more than ever the original capture is just the beginning of the process. Software will have an increasingly important role in determining the look and feel of a shot. It started with Photoshop over 30 years ago and this is another step along the path.
https://www.dpreview.com/articles/50760 ... -shock-you
Incidentally, in the comments someone mentioned a free software offering that offers advanced upscaling algorithms that is "an easy and simple to use script":
https://github.com/xinntao/Real-ESRGAN
I have not tried it so this is not a recommendation but I would be interested in feedback if anyone wants to experiment.
There is a paper that explains the principles on which it is based for anyone who really wants to get into the nitty-gritty:
https://arxiv.org/pdf/2107.10833v1.pdf
Nevertheless, I was intrigued by an analysis of DPReview about the Super Enhance feature, despite the clickbait title. I took a few things away from it. There might be an improvement to image quality by upscaling with Super Enhance then downsampling it afterwards. In the comments to the article, Barney Britton made the point that images made using cameras with Bayer sensors are already to subject to interpolation. In other words, processing commences with the development software making a best guess about the reality of what was captured. Add in camera or profile adjustments for distortion or vignetting which lens designers often employ and the "purity" principle to which some adhere begins to look hopelessly idealistic. All that really matters is what the result looks like and Super Enhance is just another tool to help reach that goal. No one should have any concerns about how it was achieved.
The article also states that Adobe has recruited Marc Levoy, an expert in computational photography, from Google. The intention is clear. Super Enhance represents a trend where more than ever the original capture is just the beginning of the process. Software will have an increasingly important role in determining the look and feel of a shot. It started with Photoshop over 30 years ago and this is another step along the path.
https://www.dpreview.com/articles/50760 ... -shock-you
Incidentally, in the comments someone mentioned a free software offering that offers advanced upscaling algorithms that is "an easy and simple to use script":
https://github.com/xinntao/Real-ESRGAN
I have not tried it so this is not a recommendation but I would be interested in feedback if anyone wants to experiment.
There is a paper that explains the principles on which it is based for anyone who really wants to get into the nitty-gritty:
https://arxiv.org/pdf/2107.10833v1.pdf