Topic: How many bits does the AD conv. of the 20D have? (Read 3993 times)

afalco · « **on:** February 28, 2005, 07:36:33 pm »

The Sony has 16Mb compressed RAW files, the Canon has 7Mb compressed RAW files, so I think it possible that the 828 has more bits in the RAW files. I'll put up some samples when I will have found out how to do that here

afalco · « **Reply #1 on:** March 01, 2005, 07:00:34 pm »

Thanks your answers! You are right it was a high contrast situation of course, but this jpeg is converted from the Cr2 file by CameraRaw 2.4. I played with the different parameters to minimize the problem, and this was the best I could achieve.
(After many attempts I couldn't find such an example that would exhibit the problem in my browser like it did in a viewer. Browsers seem to smear things out. (maybe because they just throw away the color space info?) But here's another photo (ISO 800 1/30 sec at F4). It is also converted from RAW. (I shot this in RAW only.):

Please download this detail and examine it in PS or in ACDsee or in some similar program and not in the browser!
I will learn to live together with this problem, but I really haven't seen anything like it on my F828 photos. Maybe the higher noise disguised it. And I still didn't try it out in print.

Jonathan Wienke · « **Reply #2 on:** March 04, 2005, 01:47:11 am »

Wow, what a waste. A sensor that noisy is rarely going to have more than 10 bits of actual image data; attaching a 14-bit A/D converter to a sensor that size is rather like listening to AM radio with a $10000 audiophile amp/speaker setup.

Jonathan Wienke · « **Reply #3 on:** March 04, 2005, 10:43:15 am »

Quote

It doesn't matter much in this respect but I don't think the F828 is a P&S camera (if by P&S you mean Point & Shoot :: )

Generally "P&S" or "digicam" are terms used to describe non-SLR digital cameras. Try shooting a Canon 1-series sometime and you'll begin to understand the difference.

Quote

I agree that it is strange. But what would happen if Canons also had 14 bit AD I wonder? Maybe the noise level would go up?

Assuming an ADC of equal quality, the signal/noise ratio would stay the same; the extra bits would simply be devoted to describing the sensor noise with additional precision.

Jonathan Wienke · « **Reply #4 on:** March 08, 2005, 08:37:03 pm »

I always set the shadows value to 0 in ACR, and tweak curves and levels in PS.

afalco · « **Reply #5 on:** February 28, 2005, 06:44:55 pm »

After a long time of consideration, comparisions, etc I finally collected the courage and the bucks to buy a 20D. To my great suprise I discovered that the underexposed part of my photos look like a bad scan from paper would look: irregular blotches of very few shades. It is really ugly. These blotches means that there are too few bits on the converted image for those shades. I've never noticed anything like this on photos taken with my Sony F828. (I always shoot RAW with the Canon, and - when I have enough time to wait for the Sony to save its RAW files - I shoot RAW with it too) So I started to wonder how many bits the AD converter in the 20D may have. I couldn't find any information regarding this. All I know is that the Sony has 14 bits AD converters. I suspect that the ones in the Canon have fever, maybe only 12 bits. I use the Camera RAW plugin to convert the RAW files to jpegs. Do you other Canon owners have the same problem?

Jonathan Wienke · « **Reply #6 on:** February 28, 2005, 07:59:01 pm »

The 828 uses a stupid encryption scheme (a really ill-conceived and fortunately unsuccessful effort by Sony to discourage third-party RAW converters) which expands the RAW data to 16 bits per pixel and does not allow compression. It doesn't actually have more real data bits per pixel than DSLRs. 16MB very closely corresponds to 16 bits per pixel of padded RAW data, uncompressed. I'd be willing to bet that the 828 RAWs don't vary in size by more than a few KB, while the 20D RAWs are indeed compressed and can vary in size by more than a megabyte depending on ISO settings and image content. My 1D-MkII RAWs vary from 7.1 to 10.8MB, and it has the same pixel resolution as the 20D.

Jonathan Wienke · « **Reply #7 on:** March 01, 2005, 10:16:59 pm »

That image actually has the EXIF data, which indicates you used ISO 800. And the image is using the Adobe RGB color space, which is fine, but you need to convert to sRGB any image intended to be viewed in a web browser. Both images are underexposed, especially the second one. Try shooting with the 828 at ISO 800 (if it even goes that high!) and underexpose a full stop and see what you get.

Digital Exposure And Metering Strategies

afalco · « **Reply #8 on:** March 03, 2005, 06:59:31 pm »

From here: Dpreview - Sony F828 review, Page 2 And of course from the Sony User Guide :p

afalco · « **Reply #9 on:** March 04, 2005, 08:43:18 am »

Quote

But it does beg the question that if it is cost effective for Sony to put a 14-bit AD convertor in a simple P&S, why can't Canon put one (or perhaps it has four) into the 1DsII?

It doesn't matter much in this respect but I don't think the F828 is a P&S camera (if by P&S you mean Point & Shoot :: ) But you are right that's an interesting question!

Quote

A sensor that noisy is rarely going to have more than 10 bits of actual image data...

I agree that it is strange. But what would happen if Canons also had 14 bit AD I wonder? Maybe the noise level would go up?

Quote

> These blotches
are most likely mostly due to the fact that you're using ACR for your raw conversions.

You have a good point DL! (And your WEB site is quite interesting too!) I have to make a similar test myself I think because I'm using CR V2.4 while you tested V1. I tried the standalone trial version of Capture One but I have mixed feelings about it. I am thinking about giving the free open source DCRAW a chance. I'd be glad if I could get rid of those blotches!

Jonathan Wienke · « **Reply #10 on:** February 28, 2005, 07:18:33 pm »

All Canon DSLRs have 12-bit AD converters in their sensors. I seriously doubt that the 828 actually has 14 bits of actual RAW output; DSLRs like the 20D perform much better noise-wise than small-sensor point-and-shoots like the 828, especially at high ISO settings. If your 20D is not a major improvement over the 828 in this regard, you either have a defective camera or (much more likely) something set wrong, like shooting at ISO 1600 when you meant to be at 100. Post a sample of what you're talking about and it will be a lot easier to tell what is going on.

afalco · « **Reply #11 on:** March 01, 2005, 03:17:18 pm »

I think you are right about the Sony. All of my RAW .SRF files are 17393344 bytes long. Well I learned something new again.
I try to upload an example here. Hope I succeed... What I hope you'll see is a 100% crop of an ISO 100 image unmodified.

Jonathan Wienke · « **Reply #12 on:** March 03, 2005, 11:42:34 am »

Where are you getting this Sony-has-a-14-bit-ADC idea from? AFAIK it's either 10 or 12 bits, padded to 16 bits for their bloated encrypted RAW format. 14-bit ADCs are usually reserved for $20,000+ medium format digital backs, even the $8000 Canon 1Ds-MkII is only 12-bit.

dlashier · « **Reply #13 on:** March 04, 2005, 04:34:01 am »

> These blotches

are most likely mostly due to the fact that you're using ACR for your raw conversions. This is what ACR does to my 1D shadows (and actually throughout the tonal range often).

- DL

Jonathan Wienke · « **Reply #14 on:** March 04, 2005, 07:50:12 pm »

Quote

V2.4 will show better primarily because the color noise control which defaults to 25 and also causes loss of detail.

I disagree. If you have the luminance smoothing >0 you'll lose detail, but the Color Noise Reduction does a good job of eliminating color blotches without smearing the image. I generally convert with color noise reduction set to 25 and luminance smoothing set to 0, and then run Neat Image if necessary afterwards. Neat image is better at remiving luminance noise than color noise, and ACR is the opposite. So using them in combination gives you the best features of both.

afalco · « **Reply #15 on:** March 05, 2005, 11:33:22 am »

Here is the same photo converted to 16 bit PSD by DCRAW 1.236 then saved as jpeg. I even emphasized the shadows. The blotches are gone!

afalco · « **Reply #16 on:** February 28, 2005, 07:36:00 pm »

The Sony has 16Mb compressed RAW files, the Canon has 7Mb compressed RAW files, so I think it possible that the 828 has more bits in the RAW files. I'll put up some samples when I found out how to do that here

Jonathan Wienke · « **Reply #17 on:** March 01, 2005, 04:15:32 pm »

It looks like a camera JPEG shot in high contrast conditions--the foreground is much brighter than the background. I highly recommend shooting RAW in such situations.

afalco · « **Reply #18 on:** March 03, 2005, 06:38:34 am »

Well, you are right about the Adobe RGB. I usually upload photos without any color space info ("seved for WEB") but that would modify the displayed colors. And you are right: I had to make a comparison shot. Here it is (this one is saved with sRGB color space. The whole frame is not underexposed but this bag is black and of course underexposed):

From this I conclude that the main diffeerence between the Sony and the Canon in this respect is the higher noise. and at a smaller extent the extra 2 bits from the Sony. (In your earlier post you implicitely said that the Sony does not have more bits in their RAW formats then the Canon. I was thinking about it and concluded that it is unlikely that they somehow compress the 14 bit info into 12 bits then expand to 16 bits. It would be more logical if they simply pad the highest 4 bits with 0 to get the 16 bit resolution.) So I'll keep my 20D :laugh:
But anyway: thank you for your help!

DiaAzul · « **Reply #19 on:** March 04, 2005, 04:28:29 am »

Quote

Wow, what a waste. A sensor that noisy is rarely going to have more than 10 bits of actual image data; attaching a 14-bit A/D converter to a sensor that size is rather like listening to AM radio with a $10000 audiophile amp/speaker setup.

It may be a little over kill in a Sony.

But it does beg the question that if it is cost effective for Sony to put a 14-bit AD convertor in a simple P&S, why can't Canon put one (or perhaps it has four) into the 1DsII?

Author Topic: How many bits does the AD conv. of the 20D have? (Read 3993 times)