[nick2price (OP)]

I just have a few questions on digital signal processing/image frequency which i am finding extremely difficult to understand. These include:

what is high and what is low frequency in an image?

How is amplitude of an image represented?

What is the difference between image processing and audio processing?

I am finding it really difficult to understand, as all examples i have found overwhelm me with to much complicated mathematics. So any simplified explanations would be greatly appreciated.
thanks for the help

[Vern]

I don't know how you can assign a frequency to an image. It is a series of pixels. In the old days of analogue signals a colour picture was represented by three analogue amplitudes that controlled the brightness of the three primary colours on the image.

Now days it is mostly digital where the colours are represented by numbers. The primary colours are red, blue, and green. These numbers are converted into electron beam amplitude on a video screen and the dots are caused to fluoresce by the electron beams.

The latest way is to illuminate dots on a flat screen. A matrix of conductors are arranged so that two wires address a certain dot where conductors cross. Power on the conductors light the dots. A common display is the Liquid Crystal Diode, or LCD. Another type is the plasma display which has larger dots but can be brighter.

In audio the sound amplitude is also represented by numbers these days. Digital audio has a sampling rate that determines the quality of the sound, the greater the rate, the better the sound. I'm sure there is a standard rate that is used but I don't know what it is at the moment.

[Soul Surfer]

Image processing is an extremely complex subject and it is only possible to give a very brief description in the context of these pages. It is also very mathematical indeed the moment you get beyond the most basic concepts

It is in some ways similar to and different from audio processing.   The difference is obvious if you consider a simple monaural audio signal and a simple TV image the audio signal represents the data in a single valued compression wave the image is a two dimensional array with relationships between adjacent horizontal samples and vertical samples.

Now there are two sorts of digital and image processing. One aims to analyse and recognise in some way the sounds and pictures represented. The other aims to reduce the amount of data needed to transmit a satisfactory image. This latter process is called data compression.  I propose in this reply to talk mainly about data compression because that is the process involved in most consumer equipment today.

Let's go back to a simple 625 line TV picture. A new image is transmitted once every 25th of a second and each line contains about one thousand individual elements or pixels to get a reasonable quality of picture you need at least 256 levels between dark and light This in digital terms means eight bits per pixel. So each second you need to transmit  25 (frames) x 625 lines x 1000 (pixels per line) x8 (bits) this is about 125 million bits per second if you want a colour picture multiply this by three to get the brightness of each individual colour. Basic TV samplers generate data at about one thousand million bits per second.

Now this data stream is totally excessive because it could effectively transmit a different random noise colour picture every 25th of a second. Intelligible pictures are not like this, they consist of blocks of similar colour and texture so if you can design software to process this extremely fast data stream to analyse this and replace the large data blocks with smaller ones you can significantly reduce the data rate. This is what happens in jpeg data compression (the sort of thing that is used in camera pictures. now for moving images each image has to be similar to the last and so with care you can reuse most of the data on most of the frames by just shifting it about a bit this is called Mpeg data compression and is what is used for DVDs.

To return to some of your more specific questions high frequency data in an image is fine details of texture like the grain of wood or stripes in a tartan or texture in a concrete wall. low frequency is large blocks of the same colour and brightnes or with gradual shading.

The basic amplitude of the image is the brightness from full white (which is a full complement of red green and blue) to zero which is totally black and this is represented by a number within a selected range.

[Pumblechook]

An analogue TV video signal contains all sorts of frequency components up to max which depends of the detail in the scene.  In fact our eyes see more detail in black and white so a luminance signal (brightness) is transmitted up to a much higher max frequency (bandwidth if you like) than the 2 colour (chrominance) signals.  The 3rd colour is derived by subtracting the other colours from the luminance signal.  (NTSC and PAL).

Still (jepg) and moving (mpeg) digital systems use the same feature with a full bandwidth luminance and two reduced bandwidth chrominance signals.  This is a much more (bandwidth wise) efficienct way of tranmission than 3 full bandwidth Red, Green , Blue signals.  

[lyner]

Just a few 'noddy' basics:
An audio frequency is the number of cycles per second of oscillations.
The amplitude of a picture signal corresponds to the brightness 0 for black and 1 for 'peak white', for instance. A grey would be, say 0.3.
In TV and imaging, the frequency is the 'spatial frequency', that is the number of dark / light variations (could be stripes, sometimes) per inch / cm / m. So a high spatial frequency would be present in a very fine grating but a badly focused picture of a set of very broad fence posts would contain low spatial frequencies.
Any repeating signal (pulses, squiggles, musical notes) can be represented by a series of simple tones (sinewaves). A repeated pattern (or stationary picture) can also be analysed as a set of fringes of different (spatial) frequencies.
To transmit a picture in some way down a line you need to send the changes in brightness across the picture as variations of a signal. The finer the detail, the faster changes you need to send to send the picture in a given time. This needs more Bandwidth - either a wider analogue channel or more digital bits per second.
There is an added complication with pictures because they are in two dimensions. There are horizontal and vertical frequencies associated with horizontal and vertical detail in the picture. But the principle is the same.

There is often a lot of 'redundancy' in an average picture and, once you have identified repeated patterns and areas where there is little change in brightness, you can often reduce the total information you need to send by appropriately coding the basic digital (or even analogue) informaton - as Pumblechook says.

[Pumblechook]

The technique of sending chrominance signals 'embeded' with the luminance not only meant Black and White TVs carried on working with a colour signal but the bandwidth was not increased by moving to colour.  If you were to simply transmit three colour signals you would need three times the bandwidth. I think the max frequency for luminance was about 5 to 5.5 MHz but chrominance was only about 1 MHz.  (625 line)

It is like a fine detail black and white sketch which is coloured in like a kids colouring book.. Most of the detail is in the black and white sketch but you see a colour picture.   

[erickejah]

Let's go back to a simple 625 line TV picture. A new image is transmitted once every 25th of a second and each line contains about one thousand individual elements or pixels to get a reasonable quality of picture you need at least 256 levels between dark and light This in digital terms means eight bits per pixel. So each second you need to transmit  25 (frames) x 625 lines x 1000 (pixels per line) x8 (bits) this is about 125 million bits per second if you want a colour picture multiply this by three to get the brightness of each individual colour. Basic TV samplers generate data at about one thousand million bits per second.

Is not 24 bits, 8 per each color?

[lyner]

The technique of sending chrominance signals 'embeded' with the luminance not only meant Black and White TVs carried on working with a colour signal but the bandwidth was not increased by moving to colour.  If you were to simply transmit three colour signals you would need three times the bandwidth. I think the max frequency for luminance was about 5 to 5.5 MHz but chrominance was only about 1 MHz.  (625 line)

It is like a fine detail black and white sketch which is coloured in like a kids colouring book.. Most of the detail is in the black and white sketch but you see a colour picture.   

PAL coding of colour TV signals: the greatest piece of analogue signal processing that ever was or will be, IMHO.

[Soul Surfer]

I agree it was a great piece of work but only of you use the PAL system and not the messy and innacurate US NTSC system  :-)

Erikejah  read what I said again  I started with a black and white picture and than expanded if to an 8 x 3  = 24 bits per pixcel.

[Pumblechook]

PAL is a variation of NTSC which suffers from colours shifting particularly
 over long transmission chains.   Somebody coined the term Never The Same Colour.   

If I remember correctly Sony got out of paying Telefunken royalties for their PAL system by manufacturing simpler receivers... so called Simple PAL.   The results were better than NTSC but not as good as the proper PAL.   Presumably the time ran out on the royalties and Sony then could use the full PAL system.

Just read up and Sony were effectively making NTSC receivers using the PAL signal and people thought they were buying the best quality receivers on the market. 

""In the early 1970s, Sony introduced a range of PAL Trinitron TV sets, which had a Hue control like an NTSC set. These were a breath of fresh air in comparison to the dreadful Shadow-Mask TVs of the day, and it was quite a status symbol to own one. The colour decoder contained a delay line. Telefunken sued - and lost. The dreaded Japanese had hit upon a third delay-line method, which was so devilishly simple that only someone whose brain was not saturated with pro-PAL propaganda could see it. Sony used the memoire to store a line so that it could throw away alternate lines and treat the signal as though it was NTSC*. If NTSC was as bad as it was claimed to be, Sony should have been inundated with complaints; but as it was, if you owned a Trinitron set in those days, people came round to your house to watch it with you, and the TV companies adopted the video-monitor versions as studio monitors (despite the EIA tube phosphors - it was the brightness they wanted). The irony was that the most discerning TV owners were watching PAL as NTSC. Sony changed to the PAL-D method when the Telefunken patents expired, and felt obliged to devise a hue control for that, to keep up the tradition. The control didn't do anything useful, it basically gave the user the choice of whether or not to have Hanover bars, and they dropped the idea fairly quickly.""

http://www.camerasunderwater.info/engineering/tv_stds/colortv.html

[Soul Surfer]

Yes the differences betweeen the Sony ideas and the official pal were very interesting.  they had enough technology to consider the linear gun shadomask instead of the triangle this made their tubes brighter and simpler to make as well.

A very interesting period and I was vaguely involved in it at EMI Central Rsearch labs which hleld some significant patents.

[erickejah]

I agree it was a great piece of work but only of you use the PAL system and not the messy and innacurate US NTSC system  :-)

Erikejah  read what I said again  I started with a black and white picture and than expanded if to an 8 x 3  = 24 bits per pixcel.

okay