T4i- affordable Digic 5 DSLR?

c3hammer said:
+1

I'm affraid part of the appeal of the Canon DSLR video "image" is how it's processed. Keeping what makes the Canon's so appealing with higher resolving power and eliminating moire and aliasing might be more difficult than we think.

Cheers,
Pete
Click to expand...

Hmm. I'm not sure that I understand. The Digic4 was released in 2008. A lot has changed in the expectations and standards of users, especially in regard to resolving HD video since then. The reason the video is currently resolved the way it is (as I understand it), is not because of some aesthetic reason, but rather because of the limited speed and functionality of a nearly 4 year old chip.

If Canon's new Point and Shoot cameras appear to have already done away with Moire and resolution problems by adopting the Digic5, why is there any doubt the same will be true of their DSLR line?

Sure, it might not have 4:2:2 or a full frame sensor, but we won't have to worry about things like shooting patterns anymore.
 
The "look" Canon DSLR's have given us is in no small part due to the strange way Canon implemented it's scaling and processing from 18 or 21 mega pixels down to 2 with the Digic 4 processors.

My point being we may not be able to have the best of both worlds. That buttery smooth "look" and the sharp, moire free image simultaneously. A straight up video "look" with a shallow depth of field isn't exactly what we are all hoping for, but I'm afraid may be where we're headed.

Here's the sample Mobius, from the new C300 http://vimeo.com/30215350

Here's the Mobius - Behind The Scenes clip, most likely filmed with a 5DmKII and 7D. http://vimeo.com/30215350

I personally think the "look" of the behind the scenes footage is the Canon "look" we've all come to love and the actual short filmed with the C300 is rather flat and a goofy color grade to boot :)

Anyway, now that I've made my point clear as mud, we will just have to see if the new cams will provide both sides of the coin or give up what has made the Canon's so popular to solve the issue with moire and aliasing.

Cheers,
Pete
 
I think it was in the DP Review interview with Chuck Woodfall that it was mentioned that part of the key to the 1Dx's video was the faster sensor getting the data to the Digic V+ processors fast enough for them to output clean video (Sorry, that doesn't read very well, hope you can understand it).

It is within the bounds of possibility that we'll see the 35(?) mbps IBP codec in lower end cameras but not the I-Frame, which could be up to 360 mbps (Depending on which interview you read).

And I think it was in the same interview that it was mentioned that the Digic V was built around a video precessing core.

Dave
 
what I meant by "crippled" was not that the camera would suck without 422, but that they could take this feature out for marketing reasons, even if single digic 5 was enough to deliver it (which we don't know anyway) - on the line skipping thing of current cameras, I'm not aware that we know wheter this line skipping happens at the digic4 processor, or at the sensor (i.e. maybe the sensor outputs all the data and digic4 throws most of it away, or maybe the sensor can only read so many lines fast enough for video recording) - also, the rumors guys seem to point to T4i appearing in february 2012: http://www.canonrumors.com/forum/index.php/topic,2091.0.html
 
NormanBates said:
what I meant by "crippled" was not that the camera would suck without 422, but that they could take this feature out for marketing reasons, even if single digic 5 was enough to deliver it (which we don't know anyway) - on the line skipping thing of current cameras, I'm not aware that we know wheter this line skipping happens at the digic4 processor, or at the sensor (i.e. maybe the sensor outputs all the data and digic4 throws most of it away, or maybe the sensor can only read so many lines fast enough for video recording) - also, the rumors guys seem to point to T4i appearing in february 2012: http://www.canonrumors.com/forum/index.php/topic,2091.0.html
Click to expand...

I get that, but what I'm saying is that it looks like Canon has traditionally NOT crippled features for marketing reasons. I don't think anyone expected the T2i to have the same video quality of 7D until they saw it. No one would have blamed Canon for "crippling" the low end entry-level camera. But that's not what they do, especially in regard to video features (which are something of an afterthought anyway).

As for the sensor- I'm pretty sure it just throws away the extra data because its not fast enough to process it all while encoding H.264 in real time, and that has been the Achilles's heel in the design until now. Remember, the chip came out almost 4 years ago and was designed to process pictures on sensors with around 40% less resolution than we are using now. I don't think the scaling method was chosen for aesthetic reasons, as c3hammer is implying. The Canon "look" he is describing has to do with the sensor and how it processes light and colors. Once it sends that data to the image processor, the "look" is already baked in. The processor than has to decide what to do with this data, either save it as a RAW, Jpeg, or throw out data so it can process as live view or encode to H.264. If the "look" c3hammer is describing were to do with line skipping, then it would be glaringly absent from still images since they process the entire sensor. I think the line skipping is just a matter of numbers and processor limitations.

The T3i's zoom, as I understand it, works by just throwing away a different set of pixels and using ones that are closer together. The sensor is throwing the entire image at the processor either way, it is just choosing what to work with after that point. Again, I could be wrong about any of this, but this is the impression I was given during rather technical discussions and seems to make a lot of sense the more I'm hearing.

As for 4:2:2, indeed- perhaps that is the difference between the "+" and the regular Digic V. Maybe the + has all the fancy tricks the 1Dx is being praised for (higher bitrate options, more color space, etc) and requires a special sensor precise enough to offer that information efficiently. For just eliminating moire, however, I really think all we need is an image processor fast enough to not throw out most of the data.
 
Last edited:
This is an interesting discussion for sure !!!

M. Gilden said:
...I don't think the scaling method was chosen for aesthetic reasons, as c3hammer is implying.
Click to expand...
Scaling is obviously needed to get to a 2mpixel image and they chose a method that gave them an image we would all call the Canon "look". This look is not inhearant to RAW images from the same sensor at 18 or 21 mpixels. It's all about how that RAW data is interpreted.

M. Gilden said:
...The Canon "look" he is describing has to do with the sensor and how it processes light and colors. Once it sends that data to the image processor, the "look" is already baked in.
Click to expand...
The sensor can only determine intesity of light as a voltage at a given pixel. The bayer pattern filters each pixel as either red, green and blue. The final image is made up by an interpretation of intensity of a given color for each pixel. This interpretation is the sole perview of processor routine. The sensor can only give a voltage and one of three colors. The sensor is really a pretty dumb device.

That's my take on how these things work :)

Cheers,
Pete
 
c3hammer said:
This is an interesting discussion for sure !!!
Click to expand...
Indeed, these are the kinds of discussions that make the forum great!
Scaling is obviously needed to get to a 2mpixel image and they chose a method that gave them an image we would all call the Canon "look". This look is not inhearant to RAW images from the same sensor at 18 or 21 mpixels. It's all about how that RAW data is interpreted.
Click to expand...
I was not aware that anyone referred to the video quality specifically as the "Canon look", as opposed to a camera-wide thing (stills, etc). If anything, it seemed like folks were waiting for Canon to clean up their video artifacts since they prefer the coloring, contrast and sharpness (which they referred to as the "look"). I've heard photographers go back and forth arguing about the merits of their photography camera of choice versus the competition. The Canon look versus the Nikon, versus the Sony, etc... They all get pretty passionate about it as well. And the truth is, sensors may all be based on the same technology, but they do not produce identical results. Using the same lens on a Nikon body will produce a noticeably different image than it will on a Canon, just because the sensors are physically different. Every camera has its unique "look" that a photographer sometimes prefers and builds a loyalty towards.
When I see videographers and DOPs discussing the merits of the "Canon look", I assume they are referring to the same thing- in other words, the way the colors and saturation appear out of the camera's sensor (be it still or video).

I'm open to other interpretations, though. I'm not saying you are wrong, however I am saying that I don't see much reason to assume that is the case. Your theory involves the assumption that scaling versus re sampling does something to change the "look" in a positive way (something I've never heard before), and that Canon actually realized this and did it on purpose despite introducing unwanted artifacts.

According to the rule of Occam's razor, my theory seems more probable. Its an old chip, not capable of processing the entire sensor data fast enough, and we've been stuck with it for a while.
 
my bet is that line skipping was just a quick way of getting the video feature to work; if it had unintended consequences on how much we like the image... perhaps, that weird soft-but-sharp look may be line-skipping related, but I'm not convinced: people don't seem to be complaining about C300 or Alexas because they lack line skipping, you know...

anyway, the discussion was indeed interesting, but I don't think we're advancing anymore, so please allow me to cry a few tears

a couple of weeks back I was supposed to be shooting my first short; it rained, so we had to postpone the shooting after over an hour of preparation and the first couple of takes; yesterday I used (all of) those takes to make a quick teaser:
http://vimeo.com/31756593

now, look at that horrible moire in the soldier's uniform!!
(and not in the parts I expected: I had the camo patterns tested before)

this 10 minute short is the biggest thing I've shot yet, and that is the second biggest moire issue I've ever had (the biggest one being on my video from a trip to Rome, which is moire brick hell if you're travelling with a T2i); I'll have to pay a lot more attention to this issue when we shoot for real (and I don't have an external monitor, so it won't be easy)

I will be very happy if T4i is just T3i with Digic 5 and that means no aliasing/moire and clean ISO 3200 because of proper subsampling; 422 would be icing on the cake
(edit: and that's why I think we won't get a lot more than that: it is already a bold and very welcome improvement)
 
Last edited:
A Digic Dv III is a New Processor, Not in line with a Photo Digic III , IV , Or V... Its Stricly Video. For the c300
 
Sorry, no
The Digic DVIII is an OLD processor, from before the XF line, which is why the C300 has a limited codec
 
Will Turner said:
Thats DIGIC DVIII though not DIGIC III. The two lines aren't comparable like that.
Click to expand...

if that was aimed at me: yes, I know

DIGIC DV III: old chip designed for video; to be found in the C300, and also XF300, but it's not a super pro option developed for the C300: it's been around since at least 2009 and you can find it today on $300 consumer camcorders

DIGIC III: very old chip designed for stills (no video capabilities); it was the one powering, for example, the 1D3 and the 40D from 2007
 
NormanBates said:
if that was aimed at me: yes, I know

DIGIC DV III: old chip designed for video; to be found in the C300, and also XF300, but it's not a super pro option developed for the C300: it's been around since at least 2009 and you can find it today on $300 consumer camcorders

DIGIC III: very old chip designed for stills (no video capabilities); it was the one powering, for example, the 1D3 and the 40D from 2007
Click to expand...

As long as we're being technical, the DIGIC III was very capable of standard definition video. They didn't include it on the DSLR flavors because it wasn't considered a desirable feature on pro cameras back then. But the point and shoot models were capable of roughly DVD-quality 640x480, and if I recall correctly, sometimes even at 60fps. I think this chip was the first to do that, as the Digic II cameras had resolution and/or framerates too low to rival video cameras.
I think I actually used footage taken from a old powershot years ago to compliment a DVX100a on a project (not going to lie, it wasn't high quality. But it did the job).
 
Last edited:
Take everything here with a grain of salt, but this seems to agree with our discussion:

http://www.canonrumors.com/2011/11/new-aps-c-camera-in-february/

Additionally, anyone who is still unsure of whether or not the Digic5 is really going to fix the moire problem, take a look at some of the footage coming out of the SX40 (SLR-style point and shoot camera using the Digic5):
http://www.youtube.com/watch?v=_9KXWctIW-E

Especially at around 1:10 and onward, there are scenes of trees and leaves, fur of a cat, a torn fence, lots of detail shots that look great even at youtube's 1080p compression. Some of these would have been pretty difficult to capture our our Digic4's without some noisy dancing patterns.

I'm very much looking forward to February.
 
Last edited:
psmall sensor cameras have never had such horrible aliasing/moire issues to begin with

check this SX30IS (digic 4) video, for example from 10:00 onwards:
http://vimeo.com/32107033
there is some aliasing (e.g. in the big stairs), but nearly no moire; I'd call it nearly clean, at least in comarison with my own experience in Rome (a.k.a. Brick Hell) with my T2i:
http://vimeo.com/18479345

that's why I'm still not convinced digic 5 will solve the aliasing/moire issues by itself
it would be great, though
 
NormanBates said:
psmall sensor cameras have never had such horrible aliasing/moire issues to begin with

check this SX30IS (digic 4) video, for example from 10:00 onwards:
http://vimeo.com/32107033
there is some aliasing (e.g. in the big stairs), but nearly no moire; I'd call it nearly clean, at least in comarison with my own experience in Rome (a.k.a. Brick Hell) with my T2i:
http://vimeo.com/18479345

that's why I'm still not convinced digic 5 will solve the aliasing/moire issues by itself
it would be great, though
Click to expand...

Interesting, perhaps the SX30IS did not, but I recall older Point and Shoot cameras having similar problems in movie mode. To be honest, maybe I'm dating back to the days of standard def (maybe even DigicIII), where we were scaling down 5+Mpix to 640x480 30p, and likely losing tons of data from line skipping the same way the current large sensor DSLRs are. Remember, the more of a discrepancy between the video resolution and sensor resolution, the more moire (why 720p suffers more than 1080p).

Looking around at footage of Digic4 P&S cameras, you are correct though- it seems they did not suffer from it, which makes my "proof" invalid. However, it makes sense now that I think about it. I suppose that 1080p HD coming from a smaller sensor with lower resolution to begin with would not lose as much data and naturally not suffer as much from moire anyhow. However, I still stand behind the theory- the Digic5 is supposed to be fast enough to not require line skipping anymore, but rather the ability to properly resample the entire sensor. Therefore, I see no reason to assume the problems will still be there, and so far all of the information I'm finding online seems to point in this direction.
 
and my example doesn't prove anything either: it could be that the SX30IS is doing line skipping on the digic 4 processing, but the lens (having such a long zoom range) is so soft that moire is never apparent; it's just an indication that what we see on the SX40 and other small-sensor cameras may not translate to DSLRs
 
Fixing the moire/aliasing might come at the expense of rolling shutter. That's the one thing no one is mentioning. And no, reading out of the sensor IS a limiting factor to many sensor designs, not just the circuitry that is moving it along.

Each pixel has a wire attached to it that connects to an A/D converter. The resistance/capacitance that wire sees dictates how fast that wire can be read. The slowest pixel will dictate ALL the pixels. Next, the control circuitry has to switch to the next row. Again, this takes time, just the decoding alone through the muxes will be significant. You are talking about 3 times as much switching. So that 20.45ms you see with the 7D could become 60ms.

Of course, who knows what latency the processing is adding as well, but, it's definitely not that simple as throwing a faster processor at it.

This is the reason the C300 is reading out each 1920x1080 color plane IN PARALLEL because the sensor is not fast enough to do them serially.
 
You must log in or register to reply here.
Back
Top