Another Angle on Video Compression

[mikesbytes]

Heard the statement the other day "There is no such thing as colour. Colour is the human brains interpretation of frequency"

It made me think, the current way that image is represented is by 256 steps of intensity for each of 3 frequencies (red, blue, green). However, in reality if we only have frequency, then why record the RGB value, the pixel could be represented as Frequency and Intensity, which is one less piece of information than RGB. How much saving, would depend on how many bytes need to be allocated to each value.

I'm sue there will be a valid argument against this, I'll be interested in hearing the reasoning behind it.

[creamyhorror]

In 3 bytes: 256x256x256 = 16.7m possibilities (i.e. 16.7m colours).

This is the same no matter what form you store it. Hue, Saturation, Brightness is an alternative to RGB, for example, and closer to what you mention.

[Bjs]

yep, but in reality the human eye is on average only able to tell the difference between 11,000 of them and remember these are values created in computer program language.

[mikesbytes]

Hue, Saturation & Brightness are 3 measurements.

What I'm wondering is whether the same result can be done with less bits by using 2 measurements?

[creamyhorror]

[quote:dba4a25787="mikesbytes"]
What I'm wondering is whether the same result can be done with less bits by using 2 measurements?[/quote:dba4a25787]
The number of bits determines the total number of colors you can store. 3 bytes = 24 bits = 16.7m different values. No matter how you divide the measurements (whether 8-8-8, 12-12, or another other combination), you can still only store that number of different values (colors) at maximum.
For example, going 12-12 would give you 2^12 x 2^12 = 2^24 = 16.7m possible colors.

[tomlee59]

[quote:331020f8d3="mikesbytes"]Hue, Saturation & Brightness are 3 measurements.

What I'm wondering is whether the same result can be done with less bits by using 2 measurements?[/quote:331020f8d3]

Don't confuse the number of measurements with the total number of bits required to represent them. If one measurement requires 598 bits to represent it well, there has been no victory. So simply counting up the "number of measurements" is not a particularly useful way of estimating the required number of bits to do a good job (whatever that means).

The situation is even more complex when human perception is involved. With few exceptions, video compression algorithms selectively throw away information, taking advantage of the imperfections of our senses. Exploiting these "loopholes" has yielded the impressive compression factors that we enjoy in, say, H.264.

[Gavino]

[quote:6aa23f4fdb="mikesbytes"]It made me think, the current way that image is represented is by 256 steps of intensity for each of 3 frequencies (red, blue, green). However, in reality if we only have frequency, then why record the RGB value, the pixel could be represented as Frequency and Intensity[/quote:6aa23f4fdb]
The point is each pixel does not have a single frequency, but a continuous range of frequencies, each with a different intensity. The best we can do to represent that in a digital computer is to store the intensities of a small number of frequencies. RGB works well because the human eye has three different types of 'sensor', roughly corresponding to those 3 colours.

[MJ Peg]

The frequency you're thinking of is already the Hue part of Hue, Saturation & Brightness

http://www.devx.com/projectcool/Article/19954/0/page/6
http://en.wikipedia.org/wiki/HSL_and_HSV