22 replies to this topic

[Falcon-]

A couple weeks back I had a single night to play with and used my Rebel T4i 650D to run my imaging session. I had not really planed to use the 650D as an astrocam as the the moded 350D and soon-to-be-modded 1000D serve nicely there but in this case I had the 650D on hand and did not have either of the other two cameras.

While producing a Master Bias frame in PixInsight I noticed something interesting in the usually unremarkable pixel rejection maps that the ImageIntegration process creates. I seem to have inadvertently found the locations of all the "HybridCMOS AF" focus detection pixels! They showed up in the High Pixel Rejection Map.

Looking closely I think they have done something really smart here that takes advantage of one of the "weaknesses" of bayer matrix colour imaging.

First a full frame view. I have aggressively stretched the image to make the focus points obvious. JPEG compression made things rather blurry but you get the general idea here.

(I will get to the clever bit in a few posts.....)

Attached Thumbnails

• 

[Falcon-]

Now a 100% resolution crop of an area just to the right and slightly high of center

Attached Thumbnails

• 

[Falcon-]

Now if we zoom WAY in (400%) we start to see interesting things.

Notice all the detect areas are hollow. Now why would that be....

Attached Thumbnails

• 

[Falcon-]

Finally we get to what I think is the clever part. All the previous images I had been working with the RAW bayer matrix data. If we now do a de-bayer run on this and convert it to colour we find.... It's all green!

This makes a lot of sense to me. With the RGB bayer matrix what we ACTUALLY have is RGGB. The green channel has 2x the pixels devoted to it as red or blue. So - if you MUST sacrifice any data at all from the sensor elusively using Green as the sacrificial pixel set is extremely logical. Green pixels are the only ones we have "spares" of!

BTW, the de-bayer method blurred these a bit to look like dots, but of course as you can see in the raw bayer matrix data that it is an artifact of debayer not physically true.

Attached Thumbnails

• 

[Maverick199]

Interesting. Wonder how similar this is to 550D?

[ccs_hello]

Canon starting from T4i and T5i uses these CDAF pixels in its main image sensors.
As I had stated in the past, it's good for daylight LiveView and Movie shooting but bad news on astro.

Clear Skies!

ccs_hello

[Falcon-]

Interesting. Wonder how similar this is to 550D?

It is not at all. What Canon calls "Hybrid CMOS AF" - was introduced on he T4i/650D and is also included on the new T5i/700D. The new SL1/100D has a new variant called Hybrid CMOS AF II that seems (from marketing descriptions) to be the same thing but with focus-sense pixels across 80% of the sensor area instead of only the central area.

In all other Canon DSLRs the autofocus sensors are only found in the reflex viewfinder part of the camera. This means that focus can only be achieved when the mirror is down. The cameras with Live-view can use a software method to do a slower focus metric but it is not as fast/useful as the reflex built in focus point and can only be done between exposures. The idea with these in-sensor focus points is that these cameras can do fast/accurate focus as well as continuously adjust focus while recording video.

(The 650D/700D/100D do also have the traditional AF methods in the reflex section in addition to the Hybrid CMOS AF method)

[martl]

Sounds rather bad news for me. Do you know if the new 6D also has this feature?

CS
MArtin

[Falcon-]

No, the 6D does *not* have this focus method.

[fetoma]

So have we come to the conclusion that these new hybrid sensors just put a halt to "most all" of the Canon DSLR's for imaging?

[LeCarl]

Soung VERY interesting, maybe the dark noise is a bit lower as the test say but pixels is not there (desactived for imaging) ;p

[ccs_hello]

So have we come to the conclusion that these new hybrid sensors just put a halt to "most all" of the Canon DSLR's for imaging?

Unfortunately it's going to be the new trend and it's not limited just to (the latest/greatest) Canon. I'll quote myself from another CN subforum:

RE: ASI120MM
It is using a Aptina (Micron) CMOS sensor.
IMO, the current technologies used in its sensor is not so fine-tuned toward low-light, low readout noise implementation called for by DSO imaging/viewing.

The situation may change in the near future since SONY and Aptina exchanged patent cross-licensing few months back for the entire patent portfolio. Basically Aptina will learn a lot from SONY while SONY will be granted with a few fundamental CMOS Image Sensor (CIS) original inventions (Photobit days, etc.)

One major thing that will affect future view of SONY CIS by astro folks due to this cross-licensing is:
SONY probably will produce more CMOS sensors with "contract detection" pixels embedded to perform the CDAF demanded by consumers. This means (1) some regular pixels will be replaced with CDAF pixels, (2) the gaps left by CDAF pixels will be "doctored", and (2) RAW is not true raw any more.

Again, this is mass-production day-light photography/movie vs. a very small group of niche applications. If you know SONY, the answer is obvious.

Many recent popular DSLRs with low noise characteristics are from SONY. (SONY, Nikon, Pentax, and few selected micro FourThirds !! ) IMX071 and IMX109 have not gone that route. But I predict soon dedicated CDAF pixels in main CIS will be everywhere due to competitive pressure.

Clear Skies!

ccs_hello

[Falcon-]

So have we come to the conclusion that these new hybrid sensors just put a halt to "most all" of the Canon DSLR's for imaging?

Actually I was thinking "oh look it is not as bad as we feared it might be!" in that we effectively have a 2x oversample of Green pixels vs 1x red and 1x blue - this means that in the AF areas we drop down to a 1x sample of green. The loss of spacial resolution in those specific areas is still there but given bayer matrix reconstruction blurring of spacial resolution anyway I again think "not all that bad really".

Remember - the images I posted above the result of PixInsight's Winsorized Sigma Clipping algorithm working on a stack of 134 BIAS frames and is one of the rejection-maps NOT the image itself.

Below is one of the actual bias frames (1/4000th second ISO800 "dark" frame) at 400% zoom in the center of the image (the core of the AF pixel area). Left side is RAW, right side is debayered - both heavily stretched. Notice the AF focus pixels are NOT visible at all.

Attached Thumbnails

• 

[Falcon-]

Here is a 100% crop of one of the light frames from that night (180s @ ISO800). The noise pattern looks normal, no obvious artifacts.

Attached Thumbnails

• 

[Falcon-]

Though more comprehensive testing needs to be done to REALLY answer the question it looks to me like the choice to use green pixels only instead of simply blocks of pixels greatly mitigates the loss of the focus pixels.

I would not entirely rule these cameras out of AP useage. To use an overblown example given the choice of, say, a 300D or 350D and a 650D the choice is a total no brainer. The slight pixel-level resolution loss in the center is more then made up for by the higher pixel density (higher spacial resolution on a per-pixel level), the lack of amp-glow, the low noise characteristics, live-view, USB shutter release, etc..


If there is any specific test anyone wants me to do with my 650D please let me know.

[Falcon-]

BTW - I should mention that was this done on a Mono sensor instead of a one-shot-colour/Bayer Matrix sensor the result would certainly spell doom for AP usage.

[Ent]

Another T4i data point: 100% center crop, 60s @ ISO1600

Pardon the crummy focus and LP.

[Tom and Beth]

Pardon the crummy focus and LP.

Heh, THAT looks like my type of work.

[Gary Honis]

Hi Sean,

Thanks for all the testing and analysis. When I did my comparison tests using the T4i and other Canon models, I found the T4i did better than the T3i (600D) and T2i (550D) in indoor long exposure sensitivity testing, and the T3i and T2i do not use the hybrid sensor.

For the image below, 5-minute bulb exposures were taken with a Canon 70-300 1:4-5.6 IS USM Lens at 300mm and F32 at ISO 1600. These crops were at the center of the FOV. Higher resolution test images can be seen HERE.

Attached Thumbnails

• 

[scodavis]

Guys, the Canon T4i is an excellent astro imaging camera. Please look at the following image:

http://www.cvafresno...memi.cfm?id=125

You can click on the image to view a larger version.

-Scott

[ccs_hello]

Falcon,

Can you share the first picture (full resolution) in a public site so people can map these CDAF pixels?

Thanks!

ccs_hello

[Falcon-]

Can you share the first picture (full resolution) in a public site so people can map these CDAF pixels

Done.

- The rejection-map saved as a 16bit integer FITS file can be found here. (reduced from 32bit floating point for wider compatibility). I stress again, this is not what the actual Bias or Dark frames look like but was generated by PixInsight's evaluation processes during stacking.
- For reference one of the .CR2 files from the stack can be found here.

[ccs_hello]

Thank you Falcon!