I dont really understand about noise

20droger wrote in post #5481919
The actual cause of noise is the quality (or lack of quality) of the semiconductors used in the various parts of the camera.

Cameras are a combination of analog and digital circuitry The analog circuitry is the first thing the "image" passes through. All analog circuitry has a background noise level inherent to that particular circuitry. (Think of this as the background hiss in a radio, which is where the term "noise" comes from.) Better circuitry has less noise, which is the biggest single reason that less expensive cameras have more inherent noise than the more expensive ones.

ISO (usually pronounced as three letters: I-S-O) is effectively gain. When you increase gain, you increase the level of the inherent noise in the circuitry. This is why high ISOs have more noise than low ISOs.

Various things affect the amount of inherent noise in a circuit. Temperature is probably the most obvious. The higher the temperature, the greater the noise.

Light also affects noise. The greater the average light falling on a sensor, the greater the noise level of that sensor. It is fortunate, therefore, that bright areas tend to mask noise (just as a loud passage masks the background hiss of a radio). A sensor with a high average brightness, however, will show increased noise in its darker areas.

GilesGuthrie wrote in post #5485048
Just wanted to add a couple of things to this: in simple terms, when you push the power to the sensor, you increase the chance of a false reading by each of the sensor dots. These false readings are where a sensor records red, but actually that part of the image is black (for example). This is the noise.

I pronounce it as a word: eye-zoh.

In some pictures, fine noise appearing as speckles can enhance the picture. I use Noise Ninja, and it's awesome, but as I say, there are times where I think that the noise actually contributes to the art of the shot. But if it's appearing as blotches, get rid of it.

Finally, if I were looking at your posted picture (at its posted size), I probably wouldn't be moved to run it through Noise Ninja.

Great explination guys. I always wondered why myself...thank you to the OP for asking too!

Here is another question, if they get rid of the analog part of the equation, does noise go away or is it greatly reduced? There must be a reasont hey dont do this if that were that case, right? Any idea why they would have analog circuts still?

20droger wrote in post #5481919
Light also affects noise. The greater the average light falling on a sensor, the greater the noise level of that sensor. It is fortunate, therefore, that bright areas tend to mask noise (just as a loud passage masks the background hiss of a radio). A sensor with a high average brightness, however, will show increased noise in its darker areas.

This part doesnt make sense to me though, if there is more "average light" why would you get a "greater level of noise"? Am I reading that correctly?

superdiver wrote in post #5485226
This part doesnt make sense to me though,

I don't get it either...

Noise.

Greater average light increases noise overall because the sensor is analog (not digital), and the bias on the sensor is a function of the sensor's average light. The sensor is a hugh CMOS transistor matrix with a common bias function.

When the sensor has a low average brightness, the bias is also low. This results in a lower inherent noise. Increasing the average sensor brightness requires a corresponding increase in bias. Without the increas in bias, dynamic range would go to hell in a handbasket. This automatic bias control is call agc (automatic gain control) in the radio universe. Radios work like crap without it, and so would sensors.

Sensors are analog, not digital. It is the nature of the beast. Following each cell in the sensor is an analog amplifier, mostly to establish a norm (read, establish an ISO level), followed by an analog-to-digital (a/d) converter. These are the three components in each pixel that are responsible for the noise.

If you think of a single sensor cell, it has a certain background noise level. When the light falling on that cell is high, the image level is well above the noise level. Conversely, when the light is very low, the image level for that cell may be down near or even in the inherent noise level.

Boosting the signal from that cell boosts everything, the image level and the overall inherent noise. Increasing the ISO level is boosting the signal. If the image is down near or in the noise level, boosting the ISO to the point when the image is usable may also increase the noise to the point where it, too, becomes visible. The result? Higher ISO equals more visible noise.

The amplifier also has its own inherent noise, just as audio amplifiers produce hiss with nothing coming in.

Following the ampllifier is the a/d converter. This is a part-analog/part-digital circuit. The front end of the a/d converter is itself an analog amplifier (actually, an analog comparator--an amplifier with a gain of 1). This, too, has its own inherent noise.

In a nutshell, the analog signal from each cell of the sensor is amplified to establish a norm for the sensor, then converted to digital. This gives three circuit components for each pixel that generate noise: the sensor cell, the amplifier, and the front end of the a/d converter. Once the signal is digital, the generation of noise ceases. Noise that is already present in the image, however, remains in the image.

The 1D and 1Ds series cameras have the highest quality, lowest noise analog components. The 5D and XXD cameras not quite so good, and the XXD/Rebel/Kiss cameras, not quite so good again. There's a reason for the huge 1D and 1Ds prices. (What? You thought Canon used an arbitrary pricing scheme?)

To achieve the ultra-low noise characteristics (and the extreme dynamic range and other features) of the EOS-1Ds III require components of extreme precision and stability. These components have a very steep manufacturing curve, and are extrememly expensive. The corresponding components for the EOS 40D are much easier to produce, at least by a magnitude or two, and are correspondingly less expensive. The corresponding components for the EOS Digital Rebel XSi are again much easier to produce, by at least a half to a full magnitude, and again are correspondingly less expensive.

What amazes me is not that the Rebels are so cheap, but that they are so quiet for something that cheap. Canon has really done an excellent job overall. Much better than "the other camera company."

20droger wrote in post #5488108
Greater average light increases noise overall because the sensor is analog (not digital), and the bias on the sensor is a function of the sensor's average light. The sensor is a hugh CMOS transistor matrix with a common bias function.

Oh, wait. You mean boosting the signal when you say "increase average light"?
I thought you were talking about light levels (exposure)

I was talking about the light falling on the sensor. But then, you know that.

20droger wrote in post #5491154
I was talking about the light falling on the sensor. But then, you know that.

No, I failed to understand.

20droger wrote in post #5488108
Greater average light increases noise overall because the sensor is analog (not digital), and the bias on the sensor is a function of the sensor's average light. The sensor is a hugh CMOS transistor matrix with a common bias function.

When the sensor has a low average brightness, the bias is also low. This results in a lower inherent noise. Increasing the average sensor brightness requires a corresponding increase in bias. Without the increas in bias, dynamic range would go to hell in a handbasket. This automatic bias control is call agc (automatic gain control) in the radio universe. Radios work like crap without it, and so would sensors.

Re-reading this part a few times, it makes more sense now.

I'd think that this would be the least contributing part of the three "noise adding bits" in the chain, right?

(less then higher ISO and amplifier noise )

René Damkot wrote in post #5483370
It was a good idea, it's a nice image, and the noise is not objectionable. (I don't see it at this size even )

It might show up at 100%, but I doubt it would show up in print.

My personal take is that too many of us look at an image at 100% and get worried (read anal) about noise.

With my trusty 30D, I was out yesterday afternoon doing some flower closeups - I shot at ISO 640 all afternoon to attain high shutter speeds to minimize movement from wind. It was a bright sunny day - 1/ISO @ f/16 would have worked for all of them - but at f/4 or so, I had shutter speeds in the range of 1/4000 to 1/6400 - wind movement wasn't an issue.

I've compared ISO 160 shots to ISO 640 shots - have to look at the EXIF to know what the ISO was. I pretty much find that any areas that are in the shadows will have noise no matter the ISO.

If you don't get your digital camera up to 640 to 800 plus now and then, you might as well shoot film at 400. Only cowards keep it at 100.

I agree completely. Too many people are overly concerned with having an image that is technically perfect.

I can personally attest that technical perfection is no substitute for a good eye. It doesn't even come close.

Mum2J&M,

I don't see any appreciable noise either, what I do see is an opportunity.

Print a nuber of 8x10's of that shot, get some mat borders, go to the next flea market, or craft show, set up your display and get ready to beat off every cat lover with a bat.

Terrific shot....

20droger wrote in post #5488108
Greater average light increases noise overall because the sensor is analog (not digital), and the bias on the sensor is a function of the sensor's average light. "

What you have said make perfect sense, except for this part... Like René, it confuses me.

When I read this I read (in my words, obviously the problem here...lol), is "the more light that falls on the sensor the more noise is produced", but that would be contrary to what you are explaining, or what am I missing?

Everything else makes sense, but this just seems to go against what everything else says...

Im so glad I asked the question about noise, it seems like I was not the only one confused.

I also want to say that I printed 2 images (at walmart) and the noise was fine I printed 8x10 and was really happy with the outcome-although they were a little dark, but that could be me and my unclaibrated monitor)

Mike McCusker wrote in post #5492051
Mum2J&M,

I don't see any appreciable noise either, what I do see is an opportunity.

Print a nuber of 8x10's of that shot, get some mat borders, go to the next flea market, or craft show, set up your display and get ready to beat off every cat lover with a bat.

Terrific shot....

My bad This was meant for OP...:o:o:o

Mike McCusker wrote in post #5492764
My bad This was meant for OP...:o:o:o

lol i wondered what picture you had been referring to, nice to know its mine lol thanks for the compliment

The sensor is an analog CMOS transistor matrix. As such, it has an inherent noise. The level of this noise is a function of several things, among them quality, temperature, and the average amount of light falling on the sensor.

The sensor is made of silicon. Like all silicon devices, there are good ones and not-so-good ones. The higher the quality of the sensor, the less inherent noise (all other thing being equal). Canon's sensors range from very good (Rebels), to excellent (40D), to unbelievably fantastic (1Ds III).

Silicon devices have a positive temperature coefficient in re noise. That is, the hotter they are, the noisier they are. This characteristic can easily be seen with some cheap P&S cameras. At normal temperatures, the noise level can be quite low. But leave one in the direct sunlight for a while and image quality goes to hell in a handbasket.

All CMOS transistors require a bias voltage. This bias voltage determines the portion of the transistor's operating curve that will be the center point of the signal being controlled. It is a characteristic of CMOS transistors that noise increases with the bias voltage.

Now, each transistor in the sensor is held at a constant bias level that is a function of the average signal the overall sensor is processing. With an optical sensor array, this signal is the average of the light falling upon the sensor. Therefore, noise increases as a function of the average light falling upon the sensor. This noise is, of course, most visible in those portions of the sensor that are dark, that receive less than the average amount of light. This is because those portions of the sensor receiving a greater amount of light simply override the noise with brightness.

Why is the bias a function of the average amount of light? Because a single bias voltage is applied to the entirety of the sensor. This is required, so that all cells respond alike. If each cell had its own individual bias, then some cells would be responding to light differently than other cells. This would produce unpredictable results.

What this amounts to is that the output of all cells together is averaged and used to establish the dynamic bias for the sensor for each given exposure. The bias is therefore a function of the average light falling upon the sensor.

To understand why the bias is needed, or why the bias should be dynamic instead of some predetermined static amount, requires an understanding of how analog matrices work, specifically CMOS transistor matrices. That is beyond the scope of this discussion.

As I said, there are three components that affect the noise level of a camera: the sensor, the normalization amplifiers, and the input portion of the a/d convverters.

However, each of these three are subject to their own noise considerations. With all of them, quality of manufacture is a first consideration. With all of them, temperature is a consideration. And with the sensor, the level of the bias is a consideration.

I hope this helps with the understanding of noise. But in all cases, all Canon DSLRs have very low noise, even with considerable gain (high ISOs). As you would expect, for a given set of conditions an EOS-1Ds III has orders of magnitude less noise than an EOS Digital Rebel XSi, as it well should. But that Rebel XSi is no slouch in the noise department!

Go out and take pictures, people!