Printable View
Does anybody know exactly how the Canon 7D MRAW image quality option arrives at a size of ten megapixels...is this the result of an in-camera downsizing that takes place after each capture? If so what are the effects on image quality? I am busy testing my cameras in the field but some expert opinions would be most useful
Thanks
Grant
Grant,
I will ask Roger Clark in the Digital Workflor Forum to weight in on this question,
I'd be interested in Roger's theories on this as I am under the impression that Canon has never published the methodology they use to produce the smaller RAW files. Some have speculated that these files might be less noisy than the full-sized RAW, but I guess this depends on the method used to downsize.
My understanding is that the file is demosaiced (likely the same routine the jpeg engine uses), then downsampled by a linear factor 3/4 (this varies from model to model). There is a bit of further compression if I recall correctly, in that the data is stored in a luma/chroma format with the chroma data more sparsely sampled. No tone curves or gamma are applied to the data; I'm not sure about white balance.
As far as quality, well you lose 25% in linear resolution, and the freedom to use a better demosaic than the one Canon gives you. The downsampling is done with a relatively simple algorithm, that does permit some aliasing to creep into the output data (then again, so does PS Bicubic). On the other hand the files are a little smaller than the full raw, but not in proportion to the loss in resolution because demosaic expands one color per pixel to three (some of that is gained back due to the sparser chroma sampling).
Effectively, mRAW and sRAW are approaches to image compression -- you are throwing away image information that was captured -- and once one takes that point of view the question arises whether this is a good form of image compression. There I think the answer is no; Nikon's "lossy compression" saves space without losing resolution (though I've seen example images that show it may not be properly implemented); and the video company RED uses a lossy compression (I believe ala JPEG 2000) that achieves substantial compression without loss of resolution. And neither Nikon's nor RED's approach locks you into dealing with demosaic done in-camera, precluding the possibility of using better algorithms either now or in the future. Finally, the simplest route to image compression for Canon would be to drop the charade that each pixel has 14 bits of actual image information; base ISO has less than 12 bits of image information, and by the time one gets to ISO 1600, there are less than 10 bits. The simplest route to save storage space for Canon would be to truncate the bits per pixel according to the image content.
The files are not less noisy than a full-size RAW, converted and downsampled to the mRAW dimensions. Noise is scale dependent, smaller on coarser scales; downsampling the image reveals the smaller noise at the coarser scale in the image. But downsampling is the crudest, least effective way to reduce noise, as it throws away all image data on fine scales, both noise and detail. Much better is to take the full resolution RAW and apply one of the better noise filters to it; used judiciously, this approach will reduce the fine scale noise while retaining much of the fine scale detail.Quote:
Originally Posted by John Chardine
Thanks for that response Emil, from that information it appears that there would be a quality loss associated with the MRAW option, beyond just having a ten mp version of the same image. Thanks again..Grant
Grant,
Roger is away and should return soon, I have asked him to weight in on this matter.
Hello All,
I have just returned from a week in the Maze District of Canyonlands National Park--very remote--no email, and mostly no cell phone contact. (Had rain, snow, 50+ mile an hour winds--Great trip!)
I think Emil presented a good summary, although Canon has not published a good description of their algorithm, so it is hard to evaluate exactly what is being done. In the CCD world, pixels are binned 2x2,
3x3 etc on chip so there is one readout per binned pixels, thus reducing read noise effects. But you can't do
that with Bayer sensors, and read noise is getting so low that the benefit is pretty small and the loss in resolution from "binning" is worse in my opinion for most photography (may be of use in faintest astro work).
But with a Bayer sensor, the algorithm must be a resampling/noise reduction system like Emil describes. The main reason I could see in using it is you don't intend to make big enlargements or big crops, so a lower resolution image meets your needs and you want to save card and disk space. Otherwise, I agree with Email that it is better to do the noise reduction in post processing. I don't use the reduced raw formats.
Roger
Thanks very much guys, after using the MRaw option a bit on my 7D I must say that I can't see too much difference in noise or image quality (although I only view on a laptop screen) but I was interested to just find out how they arrived at the smaller file sizes. Thanks again for all your explanations
Regards
Grant
