1200mm: Digital Teleconverter

Started 8 months ago | Discussions thread
JimH123 Senior Member • Posts: 2,748
Re: Not Interpolation Re: 1200mm: Digital Teleconverter
1

OutsideTheMatrix wrote:

Eric Nepean wrote:

OutsideTheMatrix wrote:

JakeJY wrote:

Eric Nepean wrote:

OutsideTheMatrix wrote:

But it is only throwing away data that is outside the area of interest and which may not be of any interest to the photographer. I shoot RAW+JPG now (only the JPG file part shows the DTC), why doesn't Oly make the x2 DTC filter available in Work Space so people can apply it to their RAWs? It should work the same way! Also, I sometimes use M size 5 MP for x2 DTC to get a nearly non interpolated result (5 MP vs 4 MP not much different) to get a much faster burst shooting rate, especially with the fully electronic shutter on my EM10Mk2.

But when using x2 DTC at the full 16 MP I do see more crater detail on the moon than I see without the DTC.

The RAW image contains the RGB values of every pixel. This is all the detail that was captured. Some of it is already interpolated, because every pixel is either Red or Green or Blue, and the other two colors are interpolated from adjacent pixels.

RAW contains the non-interpolated results as far as I know. It basically gives the value for each pixel and then during RAW processing is when you actually interpolate/demosaic and get the RGB values. The only pixels that may be already interpolated in the RAW file are the PDAF pixels and also pixel mapping (to fix dead or stuck pixels).

No more detail than this can be extracted from the sensor.

Example:

You have a G9, which has a 20MP sensor and no lowpass filter. You have mounted a good quality 25mm lens, and it is focussed on a wall 41 feet away.

The vertical field of view is 21 feet 3 inches (6.5m), and the sensor has 3872 pixels in the vertical direction.

Each pixel covers an area 1/16 x 1/16 inch (1.67mm x 1.67mm)

Pasted to the wall is a sheet of paper with two lines of text, printed 1/16th of an inch tall:

Donald Trump will win the USA 2020 election

Donald Trump will lose the USA 2020 election

The camera takes an image.

With 1 pixel per letter height, there is no interpolation software in the world that will bring the information in these lines back. You could just as well have written

Outside the Matrix will be the next US president

However if you add in 4x optical teleconvertor, there will be 4 pixels per per line of text, and 3-4 pixels per character, you should be able to read that.

Building a 2X or 4X Digtial TC for a RAW image isn't possible.

I don't think he is talking about necessarily doing the same as digital zoom, but rather just to do a cropped RAW. That should be theoretically possible. Camera only needs to read a smaller portion and write to the RAW file with that crop. Fuji uses a mode similar to that with a 1.25x crop to get faster burst speeds.

Yes thats why I'm thinking of using M size 5 MP and combining that with x2 DTC. Would the RAW still be 16 MP? If that's the case this might only be useful for JPG only mode for faster burst speeds as I could do M F 5 MP JPG and perhaps get over spec in fps with the electronic shutter.

A cropped RAW is certainly theoretically possible, and I think the implemmentation would not be too difficult.

To the best of my knowledge, Panasonic and and Olympus cameras don't have a feature to crop RAW images in the camera. My Sony A7III can crop from FF to APS-C size; I haven't had time to check but I think the file sizes are smaller (see )

However cropping a RAW image in post processing, should make no difference to the detail resolution compared to cropping a RAW image in the camera. Croppping in PP you get the flexibity of adjusting the crop to better suit the image content at the expense of larger image files.

But DTC is aimed the JPEG shooter.

Going back to my example, let's move the camera to 10 feet from the wall, so that the text becomes larger compared to the image vertical, which is now 62.5 inches.

Displaying the captured image at 15x20 inches, the line of text that is 1/16" high on the wall 10 feet away would be 15 thousands of an inch tall. Most people wouldn't be able to read that without a magnifying glass, and certainly not when 2 feet from the image.

The pixels on the enlarged image are still 3.9 thousands of an inch apart high, so that works out to be almost 4 pixels per character height.

JPEG processing first averages the color over 2x2 blocks of pixels. The it process 8x8 blocks of pixels and removes much of the detail.

Those characters that were 4 pixels high - now become a blur. The sensor captured the detail - JPEG threw it out.

With DTC enabled, after cropping, the JPEG block size is reduced and the amount of discarded detail is reduced, thus retaining almost as much detail as the RAW image, but over a smaller area.

hmmm thats a good point, so cropping of RAW in PP is better for proper framing of the crop, but the cropping of RAW in camera has the advantage in terms of write speed? You just explained why I like in camera DTC it is actually retaining more of the detail even in the JPG image. That's probably why there's more detail in it than what I see in JPG mode.

Now here's a different question.....okay let's say I am doing RAW+JPG L SF at 300mm on images of the moon (half moon phase so the most crater detail that you can get.) I use x2 DTC and it shows this in the JPG file. But I also have the RAW which covers four times the area of the JPG and I decide to export that from WorkSpace as a 16 bit TIFF file using the same in camera settings. Will that file show more moon crater detail than the x2 DTC L SF JPG even though the latter has 4x the magnification?

The 2x Digital TC is going to show more detail. And this is because it has figured out how to create 3 new pixels for every 1 existing pixel to make your eye see the image with more detail.

I don't have a picture of the moon to show this on. Instead, I will use a Rock Crab. Also, I am going to post process the original in Gigapixel AI to increase the image by 2x. This will represent the digital 2x image that the camera would have produced. But before I show this, keep in mind that the camera DTC is designed to work very fast and is not as good as a resizing program doing it in post processing. So this example has that advantage going for it.

Image 1: Here is the original rock crab image

Image 2: Here is the resized by 2x image done by Gigapixel AI. Actually a bit more than 2x, but lets just suppose 2x)

The second image has more visible detail. the new pixels have been mathematically added to do a best approximation of what each new pixel should be. Your eye makes use of those extra pixels to provide a convincing more detail image.

If you simply increase the size of the 1st image to match the size of the 2nd image, it won't have the new pixels and may start showing pixelization.

By the way, 2x seems to be the max size to still get good IQ. At 2x, 3 new pixels are invented for every 1 pixel in the original.

Let's say we want to expand by 4x? Now, we are going to invent 15 new pixels for every 1 existing pixel. The resizing SW ends up having to make too many guesses and the IQ just isn't as sharp.

Anyway, this example should show you that the resized image will look to have better detail because the detail will be larger and easier to see.

One more point to make. Resizing IQ is dependent upon what is being resized. If an image has not much fine detail, the resizing program has an easy time of it and can do an excellent resize and the resized image will look great. But if the image has very fine detail, the resizing is likely to make mistakes and the image can become messed up.

External SW like Topaz's Gigapixel AI can deal with fine detail better than the in-camera DTC. The background of the rock proves a good example of fine detail.

I found an image of the moon and decided to resize it with Gigapixel by 2x to show how that would look.  I don't spend much effort shooting the moon because after the first few times, it becomes boring.

 JimH123's gear list:JimH123's gear list
Olympus E-M1 Olympus E-M5 II Sony a6300 Olympus E-M1 II Sony a6500 +1 more
Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow