#1  
Old 26-03-2014, 02:57 AM
Astro_Bot's Avatar
Astro_Bot
Registered User

Astro_Bot is offline
 
Join Date: Jun 2012
Location: Brisbane
Posts: 1,605
Canon EOS 600D Pixel Mapping Experiment

Introduction

There are several conflicting sources online that describe picture and video modes of Canon EOS cameras with respect to how frames are scaled, cropped and/or interpolated from the native image size. Some sources suggest a particular mode is a 1:1 pixel crop, while other sources suggest the same mode is scaled and interpolated.

In lunar, solar and planetary photography, knowing how the image is dealt with in the camera is important for choosing the right mode in order to maximize image quality to be input to digital image processing.

The only camera I have is an EOS 600D (unmodified) so I conducted an experiment to test the various assertions made about image size and scaling. These findings may not apply to any other model, even if the mode is similarly named, since how Canon processes images internally depends on the camera's sensor, the version of DIGIC processor and firmware.

I tested the modes listed below and was able to draw two reasonably clear conclusions (well, to my satisfaction, perhaps not to yours) about suitability for colour planetary photography. There are other considerations, such as image compression and any loss of quality involved therein that I touch on, and in that respect, anything I say is only as good as my information sources and assumptions.
  • Full Size Still Image
  • HD Video
  • HD Video 3x Zoom
  • Liveview Video (via Backyard EOS)
  • Liveview Video 5x Zoom (via Backyard EOS)


Experiment Details

I set up a cardboard grid (luckily I had one ready-made that had been used for fabric cutting) to act as the target image, placed the camera on a stable tripod, hooked up a laptop with Backyard EOS 3.0 and readied my IR Remote Control. Throughout the experiment, neither the camera nor the target grid were touched at all. The lens used was a Canon 50mm f/2.8 (II) "Nifty Fifty" set at f/4, with the camera set at 1/100sec and ISO800.

After adjusting focus and exposure, I shot a series of test images or short video sequences in each of the test modes. Where video was used, I used VirtualDub to extract single frames. For each mode, I verified that each image or frame was identical to the rest.

I then measured the absolute scale displayed in X and Y axes for each mode and compared it to known facts: the reported size in pixels and the expected X:Y image ratio.

Note that, although close, the test grid was not perfectly parallel to the plane of the camera, and was not perfectly flat (this is especially visible in the full size image, at top left of picture). This proved not to be a problem as, given known precise ratios of image dimensions and the fact that only single X (width) and Y (height) measurements were requred, I was able to calculate a correction factor for the X-to-Y axis scale variance and apply it to all calculations. Measurements were taken at the horizontal and vertical centrelines where this error was minimised. After correcting for the above condition, calculations suggest an error no larger than around 0.4%, however this is not a true "error bar".

I cannot calculate distortion due to the lens, however, close examination of the images taken shows no visible distortion.


Results

Full Size Image:
5184 x 3456 pixels // X: 113.6cm // Y: 76.6cm
Calculated Ratio = 1.483:1 Expected Ratio = 1.500:1
Correction Factor (multiply Y to obtain corrected height) = 0.989
Pixel Scale: 45.63 px/cm

Obviously, the full-size image is a 1:1 pixel mapping. The maximum rate achievable in continuous shooting mode is 3.7fps up to a maximum of 34 frames when quality is set to Large Jpeg, or 6 frames when RAW is used. This may be useful in some instances, but is clearly limited in frame capture rates and must be actively managed due to limitations on the number of frames captured in any one burst.

HD Video:
1920 x 1080 pixels // X: 110.4cm // Y: 62.6cm // Corrected Y: 61.9cm
Calculated Ratio = 1.784 (compared to true 1.778), error suggested = 0.4%

Only 97.2% of full size image dimensions is shown in HD Video mode, i.e. the image is very slightly cropped and then down-sampled to 1920x1080. The sampling ratio is 1 pixel in the final frame per 2.62 pixels in the sensor.

The video is compressed with the H.264 codec in the camera before being saved to the card. The H.264 profile used in the EOS 600D is Baseline at Level 5.0, which is quite lossy. What effect this has on planetary photography video quality is unknown but some data loss is assumed. The video is saved in the Quicktime MOV container format.

Note: Original video, paused, was used to measure size as it is more representative - the original video is 1080 lines. When decoding or converting formats, blank lines or noise may be added to the bottom of the frame to produce 1088 lines (this is known as the "modulo 16 problem"). The attached Jpeg image is 1088 lines.

HD Video 3x Zoom:
1920 x 1080 pixels // X: 37.4cm // Y: 21.3cm // Corrected Y: 21.1cm
Calculated Ratio = 1.775 (compared to true 1.778), error suggested = 0.2%

Compared to the un-zoomed HD Video mode, the 3x zoom mode does not appear to be exactly one-third the size - the zoom factor is 2.95, athough this could be explained by error since it is just within the suggested error margin. The corresponding pixel area on the sensor is 1707 x 960 pixels, i.e. the image is cropped and then up-sampled to 1920x1080. The sampling ratio is 1 pixel in the final frame per 0.89 pixels in the sensor. As this last result is well outside the error suggested, it seems clear that HD Video 3x Zoom is not pixel-mapped 1:1.

Zoom may be increased up to a factor of 10x. However, no zoom level will offer 1:1 pixel mapping and up-sampling (interpolation) will worsen with increasing zoom.

As for the un-zoomed case, the video is compressed with the H.264 codec in the camera before being saved to the card.

Note: Original video, paused, was used to measure size as it is more representative - the original video is 1080 lines. When decoding or converting formats, blank lines or noise may be added to the bottom of the frame to produce 1088 lines (this is known as the "modulo 16 problem"). The attached Jpeg image is 1088 lines.

Liveview Video:
1056 x 704 pixels // X: 113.4cm // Y: 76.2cm // Corrected Y: 75.4cm
Calculated Ratio = 1.505 (compared to true 1.500), error suggested = 0.3%

Almost, but not all, the full image dimensions are shown in Liveview - 99.8% - and this is clear from the image as it does not have quite the same extent as the full size image. The image is obviously down-sampled quite dramatically for display at 1056x704. The sampling ratio is 1 pixel in the final frame per 4.90 pixels in the sensor.

The video is not compressed and is saved through Backyard EOS as RV24 (24-bit RGB frames) within an AVI container. Frame rates (up to a maximum available from the camera) appear to vary dependent on, it is suggested, exposure setting and the speed of the laptop to which the data is streamed, i.e. a faster laptop may save higher frame rates.

Liveview Video 5x Zoom:
1024 x 680 pixels // X: 22.5cm // Y: 15.2cm // Corrected Y: 15.0cm
Calculated Ratio = 1.501 (compared to true 1.500), error suggested = 0.1%

The image is obviously cropped and sampled for display. The sampling ratio is 1 pixel in the final frame per 1.003 pixels in the sensor. Given the calculated pixel area of the sensor is 1026 x 677 pixels, it is highly plausible that this mode uses a 1:1 pixel mapping (or at least is extremely close).

The video is not compressed and is saved through Backyard EOS as RV24 (24-bit RGB frames) within an AVI container. Frame rates may vary as for the un-zoomed case.


Conclusions

The Canon EOS 600D image or video mode to be used for lunar, solar or planetary imaging will depend to some extent on image scale available through the optical train and personal preference.

The full size image/continuous shooting mode may be suitable for some lunar or solar photography but is probably not suitable for planetary photography.

Given that none of the onboard HD Video modes use 1:1 or integer pixel mapping, and all compress the video with lossy H.264, the HD Video modes are not recommended for any astrophotography (even though I've used them).

For planetary targets, the Liveview 5x Zoom mode available through Backyard EOS appears to be a 1:1 pixel mapping (or at least extremely close) and produces uncompressed video. This should produce high-quality video.


----------

Attached Images (shrunk for upload): (1) Full Size (scaled by half); (2) HD Video; (3) HD Video 3x Zoom; (4) Liveview Video; (5) Liveview Video 5x Zoom.
Attached Thumbnails
Click for full-size image (FullFrame_HalfSize.jpg)
195.3 KB35 views
Click for full-size image (HDVideo.jpg)
197.6 KB29 views
Click for full-size image (HDVideo_Zoom3x.jpg)
125.0 KB29 views
Click for full-size image (LiveView.jpg)
191.1 KB28 views
Click for full-size image (LiveView_Zoom5x.jpg)
97.0 KB28 views

Last edited by Astro_Bot; 27-03-2014 at 03:31 PM. Reason: It's not real until it's edited!
Reply With Quote
  #2  
Old 26-03-2014, 03:17 AM
MrB's Avatar
MrB (Simon)
Old Man Yells at Cloud

MrB is offline
 
Join Date: Jan 2007
Location: Rockingham WA
Posts: 3,435
Very interesting read, thanks. I'm not really into planetary, but it is something I have occasionally thought about trying. This info will be very helpful.
And congrats on 1234 posts
Reply With Quote
  #3  
Old 26-03-2014, 08:09 AM
rustigsmed's Avatar
rustigsmed (Russell)
Registered User

rustigsmed is offline
 
Join Date: Mar 2012
Location: Mornington Peninsula, Australia
Posts: 3,950
nice work, i might give the 5x live view a go!
Reply With Quote
  #4  
Old 26-03-2014, 11:57 AM
Regulus's Avatar
Regulus (Trevor)
Regulus - Couer de Leon

Regulus is offline
 
Join Date: Nov 2011
Location: Devonport, Tasmania
Posts: 2,350
This was fascinating and of interest to me since I also have this camera. Thanks for posting it.

Another question I have in regard to still picture quality is how the editing software relates the cameras output pixels per inch (or cm) are translated on saving to the editors settings. Often they default to 72p/in while your stills for the Full Size Image are at 115.8p/in.
My question is; should the editors settings be changed to match the cameras output pix/in to circumvent further degradations in the image?

Trevor
Reply With Quote
  #5  
Old 26-03-2014, 03:32 PM
Astro_Bot's Avatar
Astro_Bot
Registered User

Astro_Bot is offline
 
Join Date: Jun 2012
Location: Brisbane
Posts: 1,605
Glad this is useful to someone.

In my experience, the pixels/inch number is useless until you want to print*. You have a set number of pixels; if you scale the image, you scale the total number of pixels.

You can change the pixels/inch but it makes no difference to digital image quality - it's still the same number of pixels after all. But, when you print, or use print preview, the printer will use the pixels/inch number to size the image on paper.

BTW, the Full Size image I uploaded, according to my editor, has 72px/in so I'm not sure why yours reads different.


* Some display programs might take a cue from the pixels/inch number to determine default display size, but the programs I use default to fitting the image on screen, with 100% "size" being 1:1 pixel mapping with screen resolution rather than pixels/inch.
Reply With Quote
  #6  
Old 27-03-2014, 01:44 AM
Astro_Bot's Avatar
Astro_Bot
Registered User

Astro_Bot is offline
 
Join Date: Jun 2012
Location: Brisbane
Posts: 1,605
Just had a thought. I have only the EOS 600D, so could only test one camera.

The target grid is set up in my garage, however, and there's no rush for me to take it down, so if Brisbane locals have other camera models they're happy for me to experiment with, or want to test themselves, they can bring them over - the photographic part of the experiment takes no more than half-an-hour.

PM me if interested.
Reply With Quote
  #7  
Old 03-04-2014, 08:04 PM
Astro_Bot's Avatar
Astro_Bot
Registered User

Astro_Bot is offline
 
Join Date: Jun 2012
Location: Brisbane
Posts: 1,605
Update:

A do'h! moment occured to me yesterday: I can take a 1024x680 pixel centre crop of the full size image and compare it to the Liveview 5x zoom image (1024x680) to see if they're identical. They're not - the Liveview 5x image has an ever-so-slightly larger FOV - about 2% by my reckoning.

So, Liveview 5x is extremely close to being 1:1 pixel-mapped, but not quite.
Reply With Quote
Reply

Bookmarks

Thread Tools
Rate This Thread
Rate This Thread:

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +10. The time is now 09:49 AM.

Powered by vBulletin Version 3.8.7 | Copyright ©2000 - 2024, Jelsoft Enterprises Ltd.
Advertisement
Testar
Advertisement
Bintel
Advertisement