[atsimg]http://files.abovetopsecret.com/images/member/6bc63d021fa4.jpg[/atsimg]
Like previous posts, these posts will help you if you're ...
1. New to video, and unsure how to look at at it
2. You work with video sometimes, and want to brush up or look at viewing video in a different way
3. You're just generally interested, or want to know what to look for in hoaxes
Primarily it will be about noisy images in a
post environment. This means we're not overly interested in how the noise got there, just
interpreting it and managing it once it is. Image theory applies to both stills and video photography.
This post will not be helpful if you're expecting indepth and perfect maths, text book level information, or the like. (Too much detail ruins the
point)
Already know enough about noise to be happy, or believe giving people some information for themselves will be a waste of time.
Mostly it's just some quick advice on applying practical knowledge to photography of the unknown.
Bring the Noise
Noise is a significant factor in an image. Often the images we receive aren’t taken in the most ideal conditions. Often they are under exposed,
captured in odd circumstances, or just taken with really poor camera quality. The side effect of this is often electronic (or other) noise, which gets
in the way of seeing the real detail in an image. Noise is both a pain in the booty, and a boon to anyone researching odd videos. It obscures detail
in a legitmate clip, but also reveals a hoax in many cases.
Noise reduction should be completed
before any other subsequent alterations to an image in most cases. Any enhancement of a noisy image will
also enhance the unwanted noise. In the majority of cases, if the noise cannot be reduced to an acceptable level then further work will likely be
impossible, and at very least unreliable.
If anything, this is one of the most ignored rules of image manipulation on ATS by professionals and amateurs alike.
To put this in perspective, noise in a piece of photography (video or still) can be enough to have an image rendered inadmissible as court evidence.
Here on ATS, we often see persons inadvertently pulling a ‘Hoagie’ by working with excess noise within an image, or attempting to remove noise at
an incorrect place in their work flow. Often we end up with an unsatisfyingly soft final image which gives us very little in the way of better detail.
There is, of course, the usual question of how much is too much. An ‘enhanced’ photo has been changed, and is therefore not a ‘primary’ source
… The arguments for this seem to vary between ignorant or deliberate hand waving, and lack of knowledge of the process occurring. Understanding
noise is also crucial at times to spotting forgeries, as noise often varies wildly based on camera type and lighting conditions and is over looked by
the hoaxer just as much as the hoaxed.
Regardless of our process we must always remember that our goal is to highlight or reveal information already present, not to destroy or create new
details.
Identifying Noise
The term noise is a little bit vague. It can be a lot of different things depending on the situation. In this particular instance, we will refer to
noise as the part of the signal, video or still, which is not useful in the image we’re looking at. Ideally our ‘signal to noise’ ratio should
be somewhere between good and acceptable for reduction. If our noise is larger than our signal, ie the parts of detail we want to see are completely
obscured by noise then removing the noise will do us very little good, beyond averaging out the image as we’re about to discuss.
Here is a good example of an image that, whilst we may call it noisy, the ‘noisy’ parts really have become part of the image, and attempting to
remove it would be catastrophic for most purposes:
[atsimg]http://files.abovetopsecret.com/images/member/486688db141e.jpg[/atsimg]
This is because there is no detail to reveal in the vast majority of this image. We lack both the
spatial and the
temporal resolution to
bring new details.
The usual noise generated by a camera is often modelled as ‘random Gaussian noise’. In statistics the Gaussian curve results when there is a large
number of single sources which create a single result. This means a lot of individual variation amongst pixels, with some being brighter and some
being darker. If you own a camera and would like to see the noise generated by your camera, take a snap of some evenly lit grey card and then take the
image inside your computer.
Noise can also be additive and multiplicative etc … I’m aware that ATS will have a maths expert somewhere to point this out, but for most
situations approaching noise this way works.
As stated though, we won’t just be looking at camera noise, but also some other types of noise a little later on.
edit on 16-9-2011 by Pinke because: Typo!