A first look at Canon's new version of DPP - v3.11.10.0

[Johnny]

Hi Bob!

I too found a copy floating somewhere on the Internet. I did a couple of tests and my take is that DLO does make a difference. Sometimes a huge difference. But many of the supported lenses are actually very good and does not need a lot of correction.

I think that full frame sensors in combination with certain lenses like the 17-40mm ( and other similar lenses ) will benefit from DLO. If you are making prints of a certain size I think you will see a noticable difference. How big prints will you have to make to see a clear difference? I have not come thus far yet. Maybe you will take a look at this at some point?

I also think that crop sensors, especially APS-C sensors, will benefit less from DLO in combination with EF lenses. EF-S lenses will probably benefit more.

Another new feature is "Compositing Tool". A nice feature but sluggish. My computer almost chocked.

[Bob Atkins]

Apart from DLO I suspect that functions like "compositing" (multiple exposure combinations) and HDR would be better done with external software if you have it. Within DPP the functions are somewhat limited and rather slow. On the other hand DPP is very good, can work with RAW files and is free! Og course DLO is unique to DPP and obviously can be very useful.

I agree that DLO will improve some images more than others. It will work best on lenses that aren't very good to start with, which probably means wideangle zooms in the corners of the frame. I really can't see it doing a lot for a lens like a 600/4L IS II that couldn't be done using unsharp masking. It's sharp all the way across the frame to start with!

[KeithB]

I found it interesting that no Macro lenses where supported. My EF-S 60 probably doesn't need much correction, but I want it perfect! 8^)

I wonder if they de-convolve and correct for focusing errors?

[Bob Atkins]

I'm sure they don't attempt to fix focusing errors.

I don't know how they fix the other aberrations. Could be a simple selective sharpening. It's possible they might do doing some sort of a simple deconvolution process, but I doubt it's anything very complex since those things eat up lots and lots of CPU time and even then don't work all that well for typical photographic images

In principle you can pretty much fully correct CA and distortion because neither actually throws away any information. However all the other aberrations (spherical, astigmatism, coma, field curvature) lose information that you can't really get back with a complex photographic image. You can guess at it and use things like maximum entropy methods to get the statistically best image but they don't really work all that well for real world images of 3 dimensional objects and they take a lot of time.

I'm sure Canon won't be telling us what they actually do with DLO. Their techniques are no doubt proprietary and they aren't going to be giving their trade secrets to their competitors. As long as it works, I guess we don't really need to know how it works!

[KeithB]

From your article it says:
"Adjustments are made for such aspects as spherical aberration, chromatic aberration, astigmatism, curvature of field, sagittal halo, chromatic aberration of magnification, axial chromatic aberration, diffraction, and the effects of a low-pass filter in front of the CMOS sensor."

Note that it corrects for the low pass filter! That must take some fancy processing. Are they trying to beat Nikon's D8e? It also says that it can double or triple the size of the CR2 file. What does that do to burst shooting.

[Bob Atkins]

You can't compensate for diffraction. If you could, then you wouldn't need huge aperture telescopes to resolve fine detail. You can sharpen the image a bit, but that's just making the image look a little sharper.

Similarly, I don't think you can compensate for the low pass filter. The whole purpose of the low pass filter is to throw away high frequency information which would cause aliasing effects. Once you've thrown it away, you can't get it back. Again you can sharpen the image so it looks better but the only way to get that high frequency information is to remove the low pass filter (Nikon D800E). Then of course you have to worry about aliasing (moire) effects. Processing should be able to remove some of those effects, though possible not all of them all the time.

So I'm still not inclined to believe that Canon are doing something miraculous with the DLO function. Simple non-uniform sharpening along with chromatic aberration correction could well account for all the "Adjustments" that DLO makes, or at least most of them. It's clear that DLO does indeed do some sharpening since the DPP instructions suggest setting the "sharpening" slider of the RAW adjustment window to zero rather than the normal default setting of 3.

DLO is a post exposure function in DPP, so it doesn't affect burst shooting. My guess is that DLO actually creates a new RAW file and saves it alongside the original one within the same file. That would account for the doubling of the file size. DLO is way too slow and way too computationally intensive to do in real time in the camera while shooting bursts of images, not to mention the amount of data that would have to be written to memory.

[KeithB]

So, is this just marketing speak?

While I agree on the diffraction, if you know the exact transfer function of the low pass filter, you should be able to provide a compensating gain, assuming that the detail is not lost in the noise.

[Johnny]

DLO is way too slow and way too computationally intensive to do in real time in the camera while shooting bursts of images, not to mention the amount of data that would have to be written to memory.

And the huge amount of data created by all the lens profiles. The lens profile for the 24-105mm alone was 51 megabytes.

[Bob Atkins]

I don't think you can reverse the effects of the low pass filter, even if you know how it affects the image. It blurs the image in the same sense that a Gaussian blur function blurs the image and the high frequency data is lost. Applying a Gaussian blur is easy, but there is no "Gaussian anti-blur" function that will sharpen it up again and recover the original (at least not for a complex photographic image). The same applies to the anti aliasing filter, you may know what it does, but once it's done it it's very hard to undo it when applied to a complex photographic image.

What you probably can do is come up with the optimum Unsharp Mask (or other sharpening technique) parameters that improve the sharpness of the image without introducing sharpening artifacts.

I did notice that the DLO data for the 24-105 was 51MB. That's a lot of correction data. However I'm sure it's not stored with the image. My guess is still that a corrected RAW image is stored alongside the original RAW image in the new RAW file. That way when you open the RAW image you can work on the DLO corrected version instantly rather than having to wait for it to grind through all the correction algorithms. The original must still be there because when you reopen the RAW file you can go in  and turn the DLO off then work on the original RAW file.

[Johnny]

Just wanted to share some information about a possible bug in DPP 3.11.10.0 that I came across yesterday. It looks like this version of DPP is having a rendering issue with regards to 5D3 raw files. It was mentioned by Bryan at The-Digital-Picture.

I have been looking at raw files from this camera the last few days and I though they were soft, but was thinking it had to do with the AA filter. Even low iso files look soft.

When viewed in DPP`s Edit window files will appear soft. When viewed in Quick Mode window they look sharp, as they should. When in Edit window  go to "Preferences" and set "Viewing and Saving RAW images" to "High Speed" and you will notice the difference. Noise reduction will be disabled but images will look sharp and detailed.

I have seen some of the first reviews of the camera and they are complaining about soft files. My guess is that Bryan is right about this.

Just wanted to bring this to your attention. Many are curios about the new camera.

A nice collection of raw files from the captial are available for download here: http://www.metrodcphotography.com/review/canon-5d-mark-iii-raw-sample-images-low-is/

I would say the 5D3 looks great at high iso for those who are concerned. It also looks like it is capable of rendering a lot of details.

[Bob Atkins]

Thanks for the info. I don't have a 5D3 to test DPP with, but it's pretty common for RAW files to look soft while DPP is doing whatever it does. After a short time they sharpen up. With RAW files from my 7d there's a message down at the bottom of the page that reads "Generating a high quality preview...." which appears for about 10-12 seconds.

I think there are three stages, 1st a very rough pixellated image appears instantly, then after a second or so a more detailed image appears, then after 10-12 seconds the final high resolution image is available.

I did download a 5D3 .CR2 image and it does the same thing, except you have to wait more like 20 seconds for the fuill high quality high resolution version to appear.

When you turn moire reduction off (i.e. you check "high speed"), the wait is more like 2 seconds.

So I don't think there's a problem with DPP and 5D3 files. It just takes a while to render them if you have the moire reduction turned on.

[Johnny]

Noise reduction is not working properly. I have checked files from my other cameras and they are fine. In the NR Preview windows changes are seen instantly. When I am trying Luminance NR there is no visible difference between 0-7 in the preview window or when I choose save and look at the result. I take into account that DPP needs a few second to render a high quality preview. When I move the Luma NR slider to 8 I start seeing a very small impact on the file. Luma NR is the same for setting 0-7. The normal would be to see a small impact when the slider is set to 2.

Anyway, I am glad this is not affecting other files. That would have caused me a lot of work. It affects 5D3 files only.

[Bob Atkins]

I do see the same thing. Luminance noise reduction does seem to have much less of an effect on 5D3 files than it does on 5D2 files, at least for small values of LNR. Whether this is a bug or whether the control is just has a much more subtle effect at low LNR values for the 5D3 files vs the 5D2 files I don't know.

It does look like you can effect significant noise reduction on 5D3 files, it just seems you need to move the LNR slider further to the right to do so then you do with 5D2 files. I guess as long as you can get the noise reduction you want, it's OK.

If you compare JPEG files from the RAW file with 0 LNR and 4 LNR you can see that there is s slight difference, but it is indeed very, very slight.

It's probably something Canon will need to address at some point, either to explain what's going on and why it is the way it is or to make it work on 5D3 files closer to the way it does with 5D2 files.