Theoretically, it shouldn't make a difference. The high noise in 8MP cameras is caused by the size of the pixels in the sensor. Shooting at 5MP won't increase the physical size of the pixels, so I don't imagine it would help.

In my experience the situation is polar opposite from BJL's assumption. My F828, when used in 5 megapixel mode, considerably out performs my F717 and F707. The detail is actually captured in 8 megapixel mode regardless of settings. After capture the camera downsamples to 5 megapixels (in this case) and the additional detail is not all lost in the process. Here are a couple crops. The first from my F707 and the second from the 5 megapixel mode in my F828 (downsampled in camera). Links to the full original files are below if anyone is interested in checking them.

Lin





Link to original 707 image

Link to original 828 image

Hi Shivz,

I don't really know how the A2 samples, especially for VGA or movie modes. Without reverse engineering the firmware it's only a guess - but with the burst speed that is necessary for 30 frame per second VGA capture, it's unlikely that the movie modes use the same process as still captures. Whether or not the 7 frame per second captures in the VGA mode are analogus to movie mode is another issue. Remember, all cameras capture a RAW image then convert it into one of several possible forms in either compressed or non compressed modes. Most RAW algorithms being used today are a compressed tiff file which accounts for the differentiation in file size between the RAW and uncompressed tiff modes.

It's entirely possible that the A2 may simply ignore all but the center 640x480 pixel matrix for the VGA mode, but this would seem to be counter intuitive since that would effectively constitute a "crop" and change the field of view accordingly. Since what you see in the viewfinder which is electronic TTL appears to have the identical field of view as when shooting in higher resolution modes, it would appear that the camera is still capturing with full resolution then downsampling.

Still captures on the F828 are made with full resoluton and down sampled before writing to media. This same process is used in many digicams and in most dSLR's which have multi-resolution capture modes. As for the VGA movies I haven't a clue. It would seem to be counter productive to do it this way since most are progressive scan.

Best regards,

Lin

Movie modes normally get there far greate frame rate by reading only some lines. But mild resolution drops for stil images are usually done by downsampling. For one thing, reading only selected lines would immediately cut the pixel count at least in half (reading alternate lines) to 4MP. To keep horizontal and vertical pixel counts in proportion would require going down to 2MP. So 5MP output must be done by downsampling from all pixel data.

Lost in translation!  I didn't mean Movie mode at all.  I am only talking Photography here.  A2 for example has basically a 3fps continous shooting mode.  But if you down size to VGA, 640X480 you can get up to 7fps. S

I saw this quote from Michael:

"but the reality is that the current generation of 8Mp digicams are moderately noisy. Certainly this is the case in comparison with current 6MP DSLRs and even the better 5MP digicams",

and began to wonder how one of the 8 MP cameras would perform when set to a lower resolution e.g. 5 MP?

Has anyone tried this yet? and can perhaps give some feedback?

There is a brief comment about this in Phil ASkey's review of the Sony 828 at DPReview, done by comparing its 5MP output option to output from a 5MP 2/3" camera (Sony's own 717 I think.) Not an optimal test, since perhaps the downsampling could be done better on a computer, but the conclusion was that 5MP files from the 828 were not quite as good as ones from the 5MP camera.

  One possible reason is that once the pixels shrink "too much", the percentage of the sensor area that detects light decreases significantly. So maybe the Sony 8MP 2/3" sensor is gathering significantly less information (photons) than the previous Sony 5MP 2/3" sensor, so that downsampling can reduce noise levels, but cannot fully bridge the gap.

Hi Lin, Are you positive about the down-sampling in camera? This is a pretty important bit of info.  Are you saying that if you shoot in VGA mode--the A2 has a 7fps VGA setting--that the camera still shoots in full 8mp and then resizes?  Or if the down-sampling OK only for certain sizes?

Can you back this up? Thanks, S

Lin, sorry, I was ambiguous; I was talking only about noise levels with the 828 at 5MP than with the 717, since that is what the original post aske about; you seem to be talking about the 828 giving the sharper images in that same 5MP comparion.

  Going back to Phil Askey's tests at
http://www.dpreview.com/reviews/sonydscf828/page17.asp
he, like you, gives the 828@5MP the edge on sharpness, but notes that even at 5MP, ther is a bit more visible noise insome of the 828 images.

  Actually, I would probably be happy with the trade-offs of the 8MP sensors compared to the 5's; a bit of extra NR processing when needed (many subjects seem not to need it) might well get down to the same noise levels as a 5MP 2/3" cameras while still holding the same or better detail.

Actually, the misunderstanding was probably my fault. After re-reading the original question I see that noise was really the subject.

There is a slight difference in visible noise between the five megapixel capture mode with the F828 and the native capture with the F707/F717, with the F828 having a bit more noise, but not to the point of being a major issue.

On the other hand, the detail differences are quite revealing and when 8 megapixel RAW files are converted with Adobe RAW into 5 megapixel tiffs (and later to jpg in PhotoShop or other software) the differences are even more astounding because there is none of the jpg or tiff artifacting which the F828's internal conversion algorithm introduces.

I'm considering purchasing a good resolution chart target and testing the difference between 8 and 5 megapixel downsamples from Adobe RAW conversions because the detail seems to be mostly preserved and noise is considerably reduced via the Adobe color smoothing and noise reduction features.

Best regards,

Lin

I am very curious how downsampling from sensors with "very many, rather small photosites" will perform in the long run, but evidence from you and others has me optimistic for the moment about the virtues of the approach. I like to call it "oversampling" in analogy to something done in audio to help control aliasing.

  Maybe a good way to produce files of a given resolution (pixel count, to be crude about it) that have a minimum of moire and other artifacts is to use a sensor of somewhat higher pixel count (with an anti-aliasing filter just strong enough to control moire) and then downsample. Or perhaps to reduce resolution selectively, only in regions of the image with too much visible noise, or in colour channels more than in luminance.

  One other, rather extreme, example is the observation I have read that downsampling Kodak 14n files from 14MP to 6MP gives quality that is excellent compared to other 6MP files. But hopefully, "14 to 5" downsmping is far more than enough: 8 to 5 is already quite promissing.

Hi Shivz,

I understand. The point I was trying to make is that the camera uses VGA in both movie and low resolution burst modes and that since it makes little sense to downsample the full resolution in movie mode because it's an interlaced (progressive) mode, the VGA burst mode "could" be accomplished the same way. However, if it were, then why limit it to 7 frames in a burst? The fact that the camera "can" shoot 30 frames VGA per second would strongly suggest that there is a decided difference in either image compression or image acquisition mode or there would be no 7 frame limitation.

Best regards,