It isn't so much the camera body itself (that's just a sensor) -- it's more about the lens (photographer's understanding of the task is always important as well).
Mostly what the DSLR camera body offers... is the ability to change lenses to something more optimally suited for the task.
Optimizing Field curvature
A lens focuses to different distances and we tend to think of that focused distance as a flat plane. In reality it is slightly curved. This means that when you have the best possible focus at the center, the edges will likely be just slightly less focused due to field curvature.
A true macro lens is usually optically designed to do a better job controlling field flatness and may produce a well-corrected field all the way out to the corners.
I did mention photographer's knowledge in the mix because just having the right lens isn't necessarily enough.
Optimizing Aperture Selection
Few lenses will produce their best results if the f-stop is set to shoot "wide open" (lowest possible focal-ratio that the lens can offer). You'll generally get better results by slightly stopping down the lens f-stop. This varies by lens, but is often somewhere around 2 f-stops from "wide open". (Test this for your specific lens because it does vary by lens model.)
Optimizing Focus Point Selection
If you've still got a bit of field curvature and want to improve the overall focus, move the focus point. If the center is most in focus and the corners are the most out of focus... consider selecting a point about 1/3rd of the way from the center of the frame. This is the new area of best focus. The center of the frame is no longer the best ... but it's not far away so focus is still very good here. But this trade-off now means that the corners of the frame are now much closer to optimal focus than they were previously. The overall focus quality on the entire field will improve.
Noise: (added section)
What is Noise?
There are many types of noise including thermal noise, hot-pixels, amp-glow, and even quantum noise... but the most common type of noise is read-noise.
The short answer on noise is that if you are seeing significant noise then it generally means you did not have an adequate exposure.
A very important concept to be aware of is that noise is technically not a part of exposure (despite the numerous schools that teach that it is.) Understanding this will help you appreciate what contributes to noticeable noise and how to minimize it. This topic can get a bit lengthy. It does require some detail but I'll try to avoid going into too much detail.
The digital image actually starts out as an analog image. Light travels via photons, a form of energy. These photons have wavelengths. The shorter the wavelength, the more often the wave repeats in a given distance. That means the wave carries more energy.
The photo-sensitive spots on your sensor (not technically pixels, but photo-sites) can be though of as miniature solar panels. They absorb the photon (which carries energy) and convert it to electronvolts. This is an analog process ... not a digital process.
But since it is a digital camera, it needs a way to convert the analog electronvolts into digital values. These units are often referred to as ADUs ... an acronym for Analog Digital Units. It is the ADU which we think of as the digital output from the sensor.
Consider the case where you leave the lens cap on the camera and set the camera to base ISO (usually ISO 100) and set the shutter speed to the shortest possible exposure (e.g. perhaps that's 1/4000ths for your camera). Aperture doesn't matter since the lens is capped... no light can enter the camera. You take the shot.
The camera reads-out the sensor via the same process it always does ... but in this case no actual light reached the camera. You might expect that all the "pixels" (photo-sites) should have ADU values of 0 since there was no actual light... but this isn't what happens.
The sensor requires some power just to function. The mere act of powering it up and performing a read-out without taking a shot will actually result in an image where each photo-site has some non-zero value. This is the bias level of your sensor.
More importantly ... each photo-site will not have precisely the same ADU value... they will vary just slightly. This is noise. But these levels are so low that you won't be able to see them. The image will simply appear to be black.
The noise becomes noticeable when it is amplified. If we double the value of every ADU, then average difference between them also double. If we quadruple the value of every ADU, the differences will similarly quadruple. The more we do this, the more we end up with pixels that we can see ... and they aren't all the same light level because now we can see the differences. This is the noise.
If you follow this process through, you realize these differences from photo-site to photo-site were always in the image data. They were simply too small to be noticed by a human eye. What made them noticeable was the amplification.
Why isn't ISO part of Exposure
When you change the ISO, the sensor doesn't actually change. It does not become more sensitive to light. The sensor is whatever it is. ISO is a photographic term invented by photographers to describe the sensitivity of film. In the digital world there is no such thing (we don't physically change the sensor) -- but we do keep the labeling because it is a concept that photographers understand.
The concept we call ISO (in the digital world) is actually something more accurately referred to as applied gain. There are many methods to apply the gain ... both analog methods prior to analog to digital conversation (ADC) as well as mathematical applications to the ADU values after the data goes through the ADC. Since this is process that was designed by the camera maker, the algorithm they use can vary (perhaps they do not linearly apply the gain... but use some non-linear process to amplify the shadows more strongly than they amplify the highlights (e.g. highlight-tone-priority). This creates the illusion of more dynamic range.
Regardless of what method is used to apply the gain... it is never applied until the camera shutter closes and the exposure has ended. The gain is applied as a post-exposure process. But since it is a post-exposure process ... it doesn't matter if you apply the process a fraction of a second after the exposure ended... or 10 days later. Later is later ... the key take-away is that it did not happen during the exposure and that's why ISO is not actually a part of exposure.
Noticing Noise or the Signal to Noise Ratio
The signal-to-noise ratio (or SNR) is a way of thinking about the ADUs for each photo-site in your camera.
Consider two cases... in one case we will under-expose by 6 exposure stops but will apply digital gain. In the other case we will take a correct exposure.
I'll use a made up value and state that for one particular photo-site, the bias level shows an ADU of '4'.
For my example I'll assume an f/2.8 lens but we'd like to shoot 2 stops down for sharper results ... so we'll use f/5.6.
We use a light meter and it suggests that at ISO 100 we should take a 1/2 second exposure. But we don't have a tripod (shame on us) and we need to hand-hold the shot. We'll never get a good hand-held shot at 1/2 second. We do some quick math in our heads and realize that if we boost this by 5 stops (ISO 6400) we could take this as 1/125th and we think we can hand-hold the shot at that speed.
We have some dark areas in our image which, if properly exposed, should have an ADU value of around 1024 (again... I'm making up the numbers but I picked this particularly value since it's a power of 2 so it'll make the example easier).
In the case where we take a correct exposure (ISO 100 and 1/2 second) we end up with:
Signal level of 1024
Bias level of 4
Since ISO 100 is the base ISO for the camera there is no applied gain so the SNR stands as is.
This gives us an SNR of 1024:4 ... that's not bad. We wont notice the noise in that image.
But now suppose we take the same shot but we use 1/125th at ISO 6400. This gives us:
Signal level of 16
Noise level of 4
We apply the gain (6 stops) and we get:
Signal level of 1024
Noise level of 256
16:4 isn't great ... this means the noise levels are 1/4 of the signal value ... which is terrible and we're sure to notice this much noise.
Another way to think of it is that if you increase the ISO by 6 stops, you increased the noise by 2^6 or ... you increased the noise by a factor of 64!!!
It is not entirely accurately to say that noise is caused by shooting at high-ISO. The noise is always there. The difference is that when you crank up the ISO, you hurt the signal-to-noise ratio by the same number of stops as the ISO gain.
It is the poorer signal-to-noise ratio which now makes the noise noticeable.
If you are worried about noise, you can fractionally over-expose the image (just make sure you don't over-expose so much that it results in clipping) and this will boost your SNR even higher.
Post Process Noise Reduction
At this point you should be considering that given an adequate amount of light (given a proper true exposure) most cameras will provide excellent results and noise will be difficult to notice.
But what if you do notice the noise? Noise tends to most often show up in the shadows ... not the highlights.
A second consideration is that the human eye tends to spot the noise more easily in flat areas that offer little complexity or contrast. In other words if I paint a wall completely flat but put a tiny spot of white paint on it, that spot will easily be noticed. If I paint a wall nearly white and put a tiny white spot it, that spot might be noticed... but it might escape notice. And then there's the "Where's Waldo?" case... where I have a highly complex image on the wall ... loads of complexity and strong contrasts... and I hide a tiny white spot (the same size as the others) and that spot is almost certain to escape notice from all but the most determined inspectors.
A post-processing noise-reduction technique involves creating an edge-detection mask and masking out the areas of high-complexity so that you're just left with the low-contrast (flat) areas... and just de-noise those. You can also be a bit more aggressive about de-noising in shadows and less aggressive about de-noising in highlights.
(After years of learning what noise actually is and how to deal it ... I no longer fear noise in my images and consider it to be no-big-deal.)
Why do more modern cameras seem to have less noise?
Two reasons:
- Technology has advanced a bit (but only a bit)
- They've gotten better with algorithms that cook the noise out of the image.
Many modern cameras don't produce true RAW files anymore. They produce cooked-RAW files where internal algorithms have attempted to reduce the noise before saving the file. This is a bit of a cheat. Most of the time it works in your favor ... some of the time it does not (astrophotography is an application where cooked-RAWs hurt the photographer.)
Some Conclusions on Noise
- Noise is always there and cannot realistically be eliminated (there are techniques to reduce it.)
- Noise isn't really part of exposure.
- Noise is more noticeable when you have a poor signal-to-noise ratio. It is less noticeable when you have a high signal-to-noise ratio.
- Shooting at high ISO harms the signal-to-noise ratio and makes the noise more noticeable.
- To reduce noticeable noise, shoot at base ISO and collect an adequate number of photons to create an excellent signal-to-noise ratio. This eliminates (or substantially reduces) any applied gain that needs to be added to the image and protects your SNR.
- Slightly over-expose (but make sure you don't clip the histogram) to improve the SNR
- Learn how to spot and reduce noise using post-processing tools (these are techniques you'll probably want to learn anyway regardless of how good the camera is.)
