I bought a second hand Coolpix 5000 and gave it a firmware update to give it NEF (Nikons RAW format), I began testing it, and found that I when using NEF I was getting images that were nearly as good as what I was getting with my 20D. I was impressed as this was a camera generations older, and with a much smaller sensor. I soon noticed that some of the 'noise' that I found in images was not present in the RAW generated images.
I took this test shot out the balcony door.
As is true with most compacts things are fine at 100 ISO, its when you push the boundaries a little more that things get bad. Images start to look bad when you print them. It's a sort of 'grainy look', but not quite. If you look closely at the images in image editing software you see speckles of colour. This is a segment from the 800 ISO image, so it really stands out. The grass looks like its dying in patchy areas, with green and brown bits. But its the "blotchy" purple colouring that you can see in the balcony handrail that gives this away as noise of some type.
I have a theory that what is thought of as noise in the camera generated JPEG images is not noise, but rather an artifact of the production of the JPEG in camera. Based on this, I believe that smaller compact cameras can give people better images than they are otherwise getting.
I started with splitting the channels to see what was there, and comparing this with the JPG images and noticed quite remarkable differences. From this I started to think that what really limits the compact cameras it not the fact that the sensors are smaller, but that they are let down by the image creation software.
I looked at my images, and found (looking at 100% view) that the raw image looked 'grainer', but somehow cleaner. I started to play with the settings to get the 'sharpness' to appear the same between the two images, so that I could get as close to the images that the camera generated (in tems of sharpness and tonal range)
Below I have the same segment of the image made from the NEF file, which is the sensor's RAW data hover your mouse over this image to swap the image with a segment from the camera JPG image
You may be able to noitce that the JPG image has a different appearance, to the RAW sourced one. There is an uneven-ness to the colours. There is a patchy red and green colour in the wet road, and in the grass, almost like there were dead spots. Personally I can accept the 'natural' grainy look of the RAW sourced image, as the camera ISO is set to 800 after all. It is actually much better than what you would get scanning 800 ISO 35mm film. Infact it's hardly noticable on prints I have had done.
But but its that odd colour pattern from the camera JPG that I makes prints look "gosh, that was a digital wasn't it" type of poor.
When I split the channels into RED GREEN and BLUE, I found that the camera constructed JPG was very different in character to the one constructed from RAW one I have put a segment of the red channel of each file below to demonstrate what I'm meaning. The camera JPG image is below shown first, hover your mouse over to swap with the one sourced from the RAW file.
Shocked? I was!
I think that you can see clearly that there is not just different contrasts, but that the nature and appearance of the 'noise' is different. There is a strange patterning that appears in the image, almost like interference more than noise. Further, you can see that IMAGE SHARPENING is more agressive too. Notice that the edges are harder around things and that the number 20 is more clearly visible on the building. I beleive that this shows far greater artifacts generated IN CAMERA by the system which converts the RAW data than that used in when I converted the RAW file.
Now, most digital cameras use a sensor type called a Bayer Sensor, and Bayer
sensors are significantly stronger in the represenation of GREEN pixels , so if the amount
of light is at the limits of the sensor, then its reasonable to expect that there would be
better signal available to make the GREEN than the RED when on the edge ... when
I looked at the GREEN signal from the camera generated JPG it was almost the same as the
one that came off the sensor. In the next image I have the camera JPG file first,
hover your mouse over the image to swap to RAW GREEN channel.
Nothing like as different is it!
Some more agressive sharpening is apparent, and perhaps some kind of noise reduction being applied to things too (looking at the smoothing of the grass)... but that these are so simmilar suggests to me that there is either a difference in how the camera is processing the RED channel, or that the RED channel has really bad signal (which supports the BAYER sensor argument). Yet there is not such a difference in the above GREEN as there was in the above RED.
If the difference is signal levels of RED being low, we'd expect to see this in the RAW file. If we assume that the RAW data is really what the camera processor had to work with, why did the PC Software conversion process did a much better job of converting this, with not so much difference between the GREEN and RED channel of the RAW as there was with the Camera system.
Lets review both again side by side ...
| RAW Green Channel (mouse over to see RED) | JPG Green Channel (mouse over to see RED) |
|---|---|
Not so startlingly different between the levels and appearance of RED and GREEN in the RAW is it? Yet the Camera sourced image is reminiscent of a poor static riddled TV signal while there is something less than pure in the RAW sourced channels its not horrid like the camera JPG RED is.
While I can't explain why there is so much difference between the RED in the JPG and the camera RAW, it is the striking pattern difference that makes the Camera JPG's look poor. Because to make the colour image the software has to to 'marry up' these two parts, you can see that the differences between channels will make the image uneven. The "TV static pattern" that appears in the RED doesn't match anything in the GREEN. This is where the mottled colour pattern is comming from in our final camera provided image
So, now considering this, take a look again at our two segments (hover mouse over to switch to see the JPG file) probably you can see now the "mottled" colour look is following the static pattern we saw in the RED.
I've seen this effect cited as noise, but I think that I've demonstrated here that rather than "sensor noise" it is some artifact created by the way that the data from the sensors is used to make each channel and how the channels are assembled into an image in the camera.
As well, I can see more detail in the RAW image (check the curtains in the window) and the more natural appearance of the RAW file (created by the less agressive sharpening). If nothing else, the more even colour rendition, and lack of "patchy mottled colours" on the grass and plain tones (check the wall under the window and the door of the garage) allow this image to print far more naturally than the JPG. Combined with the other advantages of RAW files (such as chromatic abberation correction) I feel that camera makers should just focus on making cameras and leave the image processing to others.
With no other changes to sensors anything on the camera, better images can be had from this (and probably all) compact digital cameras if only we had access to the RAW data. This supports that better images can be had by using RAW especially on compact cameras
While camera makers might argue that "compact camera users are not serious photographers", I think that they stand to benefit from using RAW formats in their cameras, perhaps not even having JPG conversion on the camera at all.
Considering that
Actually, this could serve to make the cameras cheaper, as the camera makers would not have develop and include proprietary embedded image processors for their cameras.
| Comparison of sensor size | Comparison of pixels from the sensor |
|---|---|
|
|
I wondered if I would see significantly better noise values in the data from the 20D. When you think that the 20D sensor is so much bigger, and generating an image not much larger than the one from the 5000. You would expect to have a far superior image, in both the JPG and the RAW file. So I thought that it might be interesting to examine what I got from the 20D. I was surprised to find that there was not such a great difference as there was when comparing RAW to in-camera JPG on a compact. My test results are here, if you want to have a look..
Consider for a moment the difference in physical sizes, they're something like this:
CP5000 sensor size = (8.80 x 6.60 mm) with an image size of 2560 x 1920 pixels
20D sensor size = (22.5 x 15.0 mm) with an image size of 3504 x 2336 pixels
The 5000 has a significantly smaller sensor producing an image that isn't that much less pixels than what comes from the 20D. Roger N Clark has quite some data on evaluating the performance of CCDs and suggests that image sensors such as the 20D are already at or near the limits of our ability to capture light. It is argued that by reducing the size of each 'pixel' on the sensor that the ability to get a noise free signal is lower.
As I mentioned above, the camera image sensor sizes are
CP5000 sensor size = (8.80 x 6.60 mm) with an image size of 2560 x 1920 pixels
20D sensor size = (22.5 x 15.0 mm) with an image size of 3504 x 2336 pixels
This gives the CP something like 290 pixels per mm of sensor and the 20D about 156 pixels per mm of sensor.
Following this 'calculation' I guesstimate that for a sensor the same size as the 20D could get (22.5 x 290) x (15 x 290 ) = 6525 X 4350 pixel area if packed to the density of the CP5000.That's about 28 Megapixels
Making a a DSLR, like this would have all the noise dis-advantages of the little Coolpix, which is unlikely to appeal to many. But I think it suggests we have not seen the 'end' of where megapixel fashion will end up. Since higher ISO is desired by many when photographing moving things (as most seem to be) then perhaps there is room to extend this less and still keep the image quality. Especially if the camera makers can fiddle more with the DATA and make it look nicer.
Already we're seeing DSLR's going to 8, 10 and 12megapixels. After all, this is really only extending the 20D dimensions by about 20%. Yet due to the misunderstandings on image size and sensor size, it will seem like camera makers are progressing in leaps and bounds to go from 6 to 12 Megapixels. When infact they're only moving around inside existing limits.
Lastly, I would like to leave you with an image made with my 5000, as an example of the sorts of noise free images that can be produced with a compact camera using RAW. I've linked this image to a full size image, which has the colour profile embedded in it, that will look better if you open it with software that supports ICC profiles. I've tried to match the two as much as possible, but I'm still sorting out how to tune an image for everyones monitor.
You might find some 'grain' if you go looking around in the full size image, but I'm sure you won't see it in a 150 or 300 DPI print. Personally I've found that it gives the image a 'clean, scanned film look'. Not bad for a compact, almost as good as from my 20D and really quite nothing like what you would get from using a JPG produced by the compact coolpix camera.
Click on the image below to download a
full sized image.
If you have any thoughts, or would like to discuss this and or my methods,
please send me an email to
I'd be glad to have a chat ...
where do you want to go now.
back to digital thoughts
back to my home page
