eXtensions - Friday 8 February 2019



advertisement


Cassandra - Topaz JPEG to RAW AI on the Mac (2): The Proof of the Pudding


By Graham K. Rogers



Topaz JPEG to RAW AI



Last week I examined an app from Topaz Labs that uses AI and neural networking to take input from JPEGs and output images in RAW (DNG) or TIFF. While the initial examination showed that the application produced what it claimed, I wanted to look at the output from the perspective of editing.


I recently downloaded a test version of Topaz JPEG to RAW AI and wrote about my initial tryout last weekend. After some early tests of converting JPEG images to TIFF and DNG (RAW), I shared some of the image output with a class of computer engineers I was teaching. One of them asked me if there really was any difference. There was, but it was not easy to discern from a quick look, apart from checking the relative file sizes of the converted images with the originals.

The question bugged me over the weekend and I went back to the images to have a closer look. That original picture I had examined in class was itself the result of a scan from a color negative I took in the early 1990s. The photograph had been taken through a glass screen, as Tony Sale and his assistant were working on the rebuilding of Colossus.

The reconstruction has a historical significance in the relationship between the origins of modern computing and now; and there is also a technical importance as the entire device was constructed from telephone technology of the 1940s and uses hundreds of vacuum tubes. The transistor, remember, was not patented until 1948 by Bill Shockley and his team at Bell Labs. I use this image often in teaching.


200% zoom
Colossus - 200% zoom


To check on the possibility of improvements, I used three images and examined the original JPEGs as well as the DNG and TIFF output from Topaz JPEG to RAW AI. As output seems not to be related directly to JPEG input size (see below) some images may fare better during editing. With the Colossus image I had to reprocess the JPEG to produce a DNG file as I had initially only created a TIFF image. That had been my first tryout of the software.

What I could see from my first look at the images was some additional crispness, and certain details around the vacuum tubes. Rotating switches on a panel, and a teletype console, were also sharper and appeared to have more detail. Likewise the camera image was sharper in the TIFF and DNG output, while the sepia image of people on a station platform also looked crisper.


200% zoom
Colossus - 1000% zoom - detail


There were slight color differences between the DNG and TIFF output in each case. DNG images showed more contrast in the darker grey areas, the sepia tone, and in the reds of the vacuum tubes. I was aware that my examination of the images was subjective and perhaps tainted by this. A more objective examination would be useful.


I opened all images using Graphic Converter (version 10.6.8). I zoomed in initially to 200%. What I had already seen with the un-enlarged images was more evident: clarity and sharpness were improved with the DNG and TIFF output. I increased the zoom up to 1000%. While details were pixelated and blocky in all parts, there was better definition in the TIFFs and even more in the DNG output, with the edges of the valves better defined.


200% zoom


With the sepia image, certain details of the face and other articles in the scene were improved. With the camera images, the improvements in lettering and icons were notable. When examining all three, the DNG of Colossus and the camera images were better, while the best sepia image output was from the TIFF image: some additional artefacts had been introduced to the face in the DNG picture although these were only noticeable when zooming to 1000%.


200% zoom


With the image of the camera, the lettering and icons of controls were far more sharp with TIFF and DNG output, and the DNG photo was improved in terms of black levels. When the images were zoomed to 1000%, there was obvious pixellation in all of them, but icons and lettering suffered less in the TIFF and DNG output, particularly in the latter.


1000% zoom


One of the reasons I prefer RAW images when taking photographs is the ability I have to edit the image and bring out details that might have been recorded digitally, but which would perhaps be lost in a JPEG rendition. While I can show that the quality of pixels is higher, the real proof of this pudding (the proof of the pudding is in the eating) would be in the editing, although this itself still has a certain subjectivity. With a larger image, and one that contains enhanced detail, editing can bring out more of what is in a RAW image, but there are always limits to what can be done, without editing at the pixel level (e.g. Photoshop) which is something I rarely do.

All of the images I produced when using Topaz JPEG to RAW AI retained the originating image size and pixel count (72dpi). I took this a little further with two of the images that had quite different input sizes. With the image of the camera (initially), I enlarged the DNG from 9" to 18" and then to 25" width (my version of Graphic Converter had been updated to 10.6.9 since the checks above). While that DNG was still capable of a clear display, the JPEG was beginning to look less viable at the larger size. Both images of the camera showed a small amount of pixellation on the thin diagonal lines of controls. The TIFF version was less sharp. The original size of the JPEG was 852KB. The TIFF and DNG outputs were 1.7MB and 1.9MB respectively.

Likewise I enlarged the sepia image from its original (277KB) 11.347" width to 22" and then to 40". Again, the DNG (5.3MB) was able to sustain a clear and sharp output, while the JPEG was not able to sustain the same enlargement. The TIFF version (5MB) showed mildly more pixellation than the DNG image.


Topaz JPEG to RAW AI
Topaz JPEG to RAW AI - output of DNG from JPEG


A larger original image will usually respond better to skillful editing: certainly to cropping. Simply enlarging an image rarely improves the picture itself: we are just using the same content and making it bigger, emphasising what already exists. The larger we go, the more the pixels are spread out. Pixellation and blockiness become more evident.

With the output from Topaz JPEG to RAW AI the larger images produced do appear to have more content included (using AI) and are able to respond far better to editing and enlargement. Perhaps we really do have something out of nothing.


See also:


Graham K. Rogers teaches at the Faculty of Engineering, Mahidol University in Thailand. He wrote in the Bangkok Post, Database supplement on IT subjects. For the last seven years of Database he wrote a column on Apple and Macs. After 3 years writing a column in the Life supplement, he is now no longer associated with the Bangkok Post. He can be followed on Twitter (@extensions_th)


advertisement



Google


Made on Mac

For further information, e-mail to

Back to eXtensions
Back to Home Page


All content copyright © G. K. Rogers 2019