3 Perceptual source-coding and lossy compression

3.1 Introduction

Source coding is the representation of a phenomenon such as a sound or an image in a form suitable for communication or storage. You are probably familiar with the example of digital sampling for creating digital representation of sounds or images. Prior to transmission, such representations often have error control incorporated, as you saw earlier. There is more to source coding than just sampling, though, because of the desirability of representing a source as efficiently as possible. Usually this means using as few bits as possible, consistent with achieving a desired level of fidelity to the original source. This raises the issue of data compression, and in particular lossy compression which I will look at shortly. Lossy compression is widely used in the source coding of sound, images and video. MP3 and JPEG files use lossy compression. Usually this compression relies on human perceptual characteristics that enable some data to be discarded with no apparent degradation.

In this introductory audio, Allan Jones talks to Laurence Dooley, the author of Section 3, about some of the issues from this section.

Download this video clip.Video player: tm355_openlearn_003-640x360.mp4

Show transcript|Hide transcript

Transcript

ALLAN JONES

Hello again. I'm Allan Jones, and I'm talking now to Laurence Dooley, who produced the material for section 3 of this course. Hello, Laurence.

LAURENCE DOOLEY

Hello.

ALLAN JONES

You're concerned in section 3 with lossy compression, which is a form of source coding. But there's another form of source coding which you don't deal with, called lossless compression. What's the difference between lossless compression and lossy compression?

LAURENCE DOOLEY

With lossless compression, the process is reversible. So you're guaranteed to get an exact version of the original source, so you get back what you had before you applied any compression, just as you would with, for instance, a zip file. The same is not true with lossy compression. Some details in the source are irredeemably lost during the compression process, and the reconstructed version will only be an approximation of the original source, albeit, often, it is interpreted by us humans as being exactly the same, like in MP3 or JPEG.

ALLAN JONES

Lossy compression sounds like the poor relation. It sounds as though using lossy compression is inferior to lossless compression because it discards information.

LAURENCE DOOLEY

Yes. It's true that lossy compression discards information, but I certainly wouldn't say that it's inferior. The trick with lossy compression is that what is lost or discarded are things that are not perceived. For multimedia type signals, the way we humans interpret sounds or images can be exploited in the compression process. For example, in audio coding, we can afford to lose sounds that we can't hear, while in image and video coding, we can discard things we're unable to see.

So even though some detail is lost, hence the term "lossy coding," for us humans, there's no change in what we perceive, either audibly or visually. This approach to source compression is known as perceptual coding. This is distinct from the data coding you encountered in lossless compression, because rather than only focusing on exploiting data patterns or long strings of binary digits, we're interested in our perception of the source.

ALLAN JONES

OK. So why would you bother? Why not use lossless compression?

LAURENCE DOOLEY

Well, it's all to do data, bit rates, and file sizes. Lossy methods can compress files much more than lossless compression. They also offer much greater flexibility, as there's now an inherent trade-off existing between file size or transmission bit rate and the perceived quality. We can accept some reduction in the perceived quality in exchange for higher compression ratios or smaller file sizes.

ALLAN JONES

OK. Can you say something about how lossy coding is actually done?

LAURENCE DOOLEY

Yes. Very often, you find the part of the process involves a change of domain. For instance, some of the processing involved in creating MP3 files is done in the frequency domain, to identify those audio frequencies which we perceive and those we don't, and consequently, those we don't need to bother encoding.

The same is true for image coding, like JPEG. The part of the processing is done in what we call the spatial domain, where we now consider spatial frequencies, a concept which is not so straightforward to grasp. But if you remember that sine waves in space can have different wavelengths, then imagine that the spatial variations in image can be resolved into sine waves having different wavelengths.

The key idea to remember about part 3 is that of perceptual coding, which as I mentioned before, is especially applicable to multimedia content like audio, images, and video. If the person at the end of a communications link is not going to be able to either hear or see something, then why bother coding and transmitting this information?

ALLAN JONES

Thanks.

 

End transcript

Download