The document discusses the process of digitization. It explains that digitization involves sampling an analog signal at discrete time intervals and quantizing the sampled values to a fixed set of levels. This converts a continuously varying analog signal into a digital signal represented by discrete numbers. Sampling reduces the continuous signal to a sequence of equally spaced values, while quantization restricts the values to allowed levels represented in a fixed number of bits. Together, sampling and quantization allow analog signals from physical phenomena to be represented digitally.
30. 22 F U N DA M E N TA L S
In multimedia, we encounter values of several kinds that change continuously, either because
Digitiz ation
they originate in physical phenomena or because they exist in some analogue representation.
For example, the amplitude (volume) of a sound wave varies continuously over time, as does the
amplitude of an electrical signal produced by a microphone in response to a sound wave. The
colour of the image formed inside a camera by its lens varies continuously across the image plane.
As you see, we may be measuring different quantities, and they may be varying either over time
or over space (or perhaps, as in the case of moving pictures, both). For this general discussion, we
will follow tradition, and refer to the value we are measuring, whatever it may be, as a “signal”,
not usually distinguishing between time-varying and space-varying signals.
When we have a continuously varying signal, such as the one shown
in Figure 2.2, both the value we measure, and the intervals at which
we can measure it, can vary infinitesimally. In contrast, if we were to
convert it to a digital signal, we would have to restrict both of these to
a set of discrete values that could be represented in some fixed number
of bits. That is, digitization – the process of converting a signal from
analogue to digital form – consists of two steps: sampling, when we
measure the signal’s value at discrete intervals, and quantization, when
Figure 2.2. An analogue signal
we restrict the value to a fixed set of levels. Sampling and quantization
can be carried out in either order; Figure 2.3 shows a signal being
ÊสÑัÞญÞญÒา³ณ first sampled and then quantized. In the sampling step, you see the
analogue continuous signal reduced to a sequence of equally spaced values; in the
quantization step, some of these values are chopped off so that every
one lies on one of the lines defining the allowed levels.
These processes are normally carried out by special hardware devices, 16
31. 22 F U N DA M E N TA L S
In multimedia, we encounter values of several kinds that change continuously, either because
Digitiz ation
they originate in physical phenomena or because they exist in some analogue representation.
For example, the amplitude (volume) of a sound wave varies continuously over time, as does the
amplitude of an electrical signal produced by a microphone in response to a sound wave. The
colour of the image formed inside a camera by its lens varies continuously across the image plane.
As you see, we may be measuring different quantities, and they may be varying either over time
or over space (or perhaps, as in the case of moving pictures, both). For this general discussion, we
will follow tradition, and refer to the value we are measuring, whatever it may be, as a “signal”,
¡กÃรÐะºบÇว¹น¡กÒาÃร Analogue to Digital Converter (ADC)
not usually distinguishing between time-varying and space-varying signals.
When we have a continuously varying signal, such as the one shown
in Figure 2.2, both the value we measure, and the intervals at which
we can measure it, can vary infinitesimally. In contrast, if we were to
convert it to a digital signal, we would have to restrict both of these to
a set of discrete values that could be represented in some fixed number
of bits. That is, digitization – the process of converting a signal from
analogue to digital form – consists of two steps: sampling, when we
measure the signal’s value at discrete intervals, and quantization, when
Figure 2.2. An analogue signal
we restrict the value to a fixed set of levels. Sampling and quantization
can be carried out in either order; Figure 2.3 shows a signal being
ÊสÑัÞญÞญÒา³ณ first sampled and then quantized. In the sampling step, you see the
analogue continuous signal reduced to a sequence of equally spaced values; in the
quantization step, some of these values are chopped off so that every
one lies on one of the lines defining the allowed levels.
These processes are normally carried out by special hardware devices, 16
32. 22 F U N DA M E N TA L S not usually distinguishing between time-varying and space-varying sig
When we have a continuously varying sign
In multimedia, we encounter values of several kinds that change continuously, either the value we measure,
in Figure 2.2, both because
Digitiz ation
they originate in physical phenomena or because they exist in some can measure it, can vary infinitesimally.
we analogue representation.
For example, the amplitude (volume) of a sound wave varies continuously over time, as does thewe would have
convert it to a digital signal,
amplitude of an electrical signal produced by a microphone in response to a sound wave. The
a set of discrete values that could be represen
colour of the image formed inside a camera by its lens varies continuously across the digitization – the process o
of bits. That is, image plane.
As you see, we may be measuring different quantities, and they mayanalogue to either over timeconsists of two
be varying digital form –
or over space (or perhaps, as in the case of moving pictures, both). For this general discussion, we discrete interva
measure the signal’s value at
will follow tradition, and refer Figure valueAn analogue signal whatever it maythe value“signal”, set of levels.
to the 2.2. we are measuring, we restrict be, as a to a fixed
¡กÃรÐะºบÇว¹น¡กÒาÃร Analogue to Digital Converter (ADC)
not usually distinguishing between time-varying and space-varying signals. carried out in either order; Figure
can be
first sampled and then quantized. In the s
When we have a continuously varying signal, such as the one shown a sequence of e
continuous signal reduced to
in Figure 2.2, both the value we measure, and the intervalssome of these values are
quantization step, at which
we can measure it, can vary infinitesimally. Inlies on one of the lines defining the allo
one contrast, if we were to
convert it to a digital signal, we would have to restrict both of these to
a set of discrete values that could be represented in some fixed normally carried out b
These processes are number
of bits. That is, digitization – the processcalled analogue atosignal from
of converting digital converters (ADCs
analogue to digital form – consists of two steps:not examine. We will only consider
we will sampling, when we
measure the signal’s value at discrete intervals, and quantization, whensuccessive samp
where the interval between
Figure 2.2. An analogue signal
we restrict the value to a fixed set of levels. Samplingaand quantization time or space
samples in fixed amount of
can be carried out in either order; Figure 2.3 shows we will generally assume that
rate. Similarly, a signal being
ÊสÑัÞญÞญÒา³ณ first sampled and then quantized. In theissampling step, you see the levels – are
sampling quantized – the quantization
analogue continuous signal reduced to a sequence of equally spaced values; in the
quantization step, some of these values are chopped off so that every that digital
One of the great advantages
one lies on one of the lines defining the allowed levels. stems from the fact that on
analogue ones
those at the quantization levels – are valid
These processes are normally carried out by special hardware devices, 16
33. 22 F U N DA M E N TA L S not usually distinguishing between time-varying and space-varying bit
of sig
analog
When we have a continuously varying sign measu
Figure 2.2. An analogue signal
In multimedia, we encounter values of several kinds that change continuously, either the value we measure,
in Figure 2.2, both because we re
Digitiz ation
they originate in physical phenomena or because they exist in some can measure it, can vary infinitesimally.
we analogue representation.
For example, the amplitude (volume) of a sound wave varies continuously over time, as does thewe would first s
convert it to a digital signal,
can b
have
amplitude of an electrical signal produced by a microphone in response to a sound wave. The contin
a set of discrete values that could be represen
quant
colour of the image formed inside a camera by its lens varies continuously across the digitization – the process o
image plane.
of bits. That is, one li
As you see, we may be measuring different quantities, and they mayanalogue to either over timeconsists of two
be varying digital form –
or over space (or perhaps, as in the case of moving pictures, both). For this general discussion, we discrete interva
measure the signal’s value at These
will follow tradition, and refer to the
Figure value we are measuring, whatever it may be, as a “signal”,
2.2. An analogue signal
¡กÃรÐะºบÇว¹น¡กÒาÃร Analogue to Digital Converter (ADC)
we restrict the value to a fixed set of levels.
called
not usually distinguishing between time-varying and space-varying signals. carried out in either order; Figure
can be we w
first sampled and then quantized. In where the s
When we have a continuously varying signal, such as the one shown a sequence of e
continuous signal reduced to sampl
in Figure 2.2, both the value we measure, and the intervalssome of these values areS
quantization step, at which rate.
we can measure it, can vary infinitesimally. Inlies on one of the lines defining the qua
one contrast, if we were to
is
allo
convert it to a digital signal, we would have to restrict both of these to
One
a set of discrete values that could be represented in some fixed normally carried out b
These processes are number
analog
of bits. That is, digitization – the processcalled analogue atosignal from
of converting digital converters (ADCs
those
analogue to digital form – consists of two steps:not examine. We will only consider a
we will sampling, when we over
measure the signal’s value at discrete intervals, and quantization, whensuccessive inevit
where the interval between samp
Figure 2.2. An analogue signal Figure 2.3. Sampling and
we restrict the value to a fixed set of levels. Samplingaand quantization time or space
quantization
samples in fixed amount of ence
can be carried out in either order; Figure 2.3 shows we will generally assume that
rate. Similarly, a signal being
ÊสÑัÞญÞญÒา³ณ first sampled and then quantized. In theissampling step, you see the levels – are
sampling quantization
quantizedapisake@gmail.com.................
Ex Libris – the quantization
analogue continuous signal reduced to a sequence of equally spaced values; in the
quantization step, some of these values are chopped off so that every that digital
One of the great advantages
one lies on one of the lines defining the allowed levels. stems from the fact that on
analogue ones
those at the quantization levels – are valid
These processes are normally carried out by special hardware devices, 16
34. 22 F U N DA M E N TA L S not usually distinguishing between time-varying and space-varying bit
of sig
analog
When we have a continuously varying sign measu
Figure 2.2. An analogue signal
In multimedia, we encounter values of several kinds that change continuously, either the value we measure,
in Figure 2.2, both because
ÊสÑัÞญÞญÒา³ณ´ดÔิ¨จÔิ·ท∙ÑัÅล¨จÐะ
we re
Digitiz ation
they originate in physical phenomena or because they exist in some can measure it, can vary infinitesimally.
we analogue representation. can b
ºบÃรÃรàเ·ท∙Òา»ป˜˜ÞญËหÒาàเÃร×ื่Íอ§ง
For example, the amplitude (volume) of a sound wave varies continuously over time, as does thewe would first s
convert it to a digital signal, have
amplitude of an electrical signal produced by a microphone in response to a sound wave. The noise contin
a set of discrete values that could be represen
quant
colour of the image formed inside a camera by its lens varies continuously across the digitization – the process o
image plane.
of bits. That is, one li
As you see, we may be measuring different quantities, and they mayanalogue to either over timeconsists of two
be varying digital form –
or over space (or perhaps, as in the case of moving pictures, both). For this general discussion, we discrete interva
measure the signal’s value at These
will follow tradition, and refer to the
Figure value we are measuring, whatever it may be, as a “signal”,
2.2. An analogue signal
¡กÃรÐะºบÇว¹น¡กÒาÃร Analogue to Digital Converter (ADC)
we restrict the value to a fixed set of levels.
called
not usually distinguishing between time-varying and space-varying signals. carried out in either order; Figure
can be we w
first sampled and then quantized. In where the s
When we have a continuously varying signal, such as the one shown a sequence of e
continuous signal reduced to sampl
in Figure 2.2, both the value we measure, and the intervalssome of these values areS
quantization step, at which rate.
we can measure it, can vary infinitesimally. Inlies on one of the lines defining the qua
one contrast, if we were to
is
allo
convert it to a digital signal, we would have to restrict both of these to
One
a set of discrete values that could be represented in some fixed normally carried out b
These processes are number
analog
of bits. That is, digitization – the processcalled analogue atosignal from
of converting digital converters (ADCs
those
analogue to digital form – consists of two steps:not examine. We will only consider a
we will sampling, when we over
measure the signal’s value at discrete intervals, and quantization, whensuccessive inevit
where the interval between samp
Figure 2.2. An analogue signal Figure 2.3. Sampling and
we restrict the value to a fixed set of levels. Samplingaand quantization time or space
quantization
samples in fixed amount of ence
can be carried out in either order; Figure 2.3 shows we will generally assume that
rate. Similarly, a signal being
ÊสÑัÞญÞญÒา³ณ first sampled and then quantized. In theissampling step, you see the levels – are
sampling quantization
quantizedapisake@gmail.com.................
Ex Libris – the quantization
analogue continuous signal reduced to a sequence of equally spaced values; in the
quantization step, some of these values are chopped off so that every that digital
One of the great advantages
one lies on one of the lines defining the allowed levels. stems from the fact that on
analogue ones
those at the quantization levels – are valid
These processes are normally carried out by special hardware devices, 16
38. Digitization
24 F U N DA M E N TA L S
the same level. The effects of such undersampling on the way in which
the reconstructed signal will be perceived depend on what the signal
represents – sound, image, and so on – and whether it is time-varying
or space-varying. We will describe specific instances later. Suffice it to
say, for now, that they are manifested as distortions and artefacts which
are always undesirable.
si si+1 It is easy enough to see that if the sampling rate is too low some detail
will be lost in the sampling. It is less easy to see whether there is ever
Figure 2.5. Undersampling any rate at which we can be sure that the samples are close enough
together to allow the signal to be accurately reconstructed, and if
there is, how close is close enough. To get a better understanding of
undersampling these matters, we need to consider an alternative way of representing
a signal. Later, this will also help us to understand some related aspects
of sound and image processing.
You are probably familiar with the idea that a musical note played on
an instrument consists of waveforms of several different frequencies
added together. There is the fundamental, which is the pitch associated 18
39. Digitization
24 F U N DA M E N TA L S
¡กÒาÃร undersampl
the same level. The effects of such undersampling
ingon¢ขthe §งáแËหin which
Íอ way Åล‹‹§ง
¡กÓำàเ¹นÔิ´ด ¡ก‹‹Íอ perceived depend on
the reconstructed signal will be ãใËหŒŒàเ¡กÔิ´ด»ป˜˜ÞญËห what the signal
r– construct ÊสÑัÞญ
represents esound, image, and so on – and whether is
Òาitãใ¹นtime-varying
¡กÒาÃร
Þญ instances later. Suffice Åล
or space-varying. We will describe specific Òา³ณ·ท∙Õี่¨จÐะáแÊส´ด§ง it to
¼ผ
say, for now, that they are manifested as distortions and artefacts which
are always undesirable.
si si+1 It is easy enough to see that if the sampling rate is too low some detail
will be lost in the sampling. It is less easy to see whether there is ever
Figure 2.5. Undersampling any rate at which we can be sure that the samples are close enough
together to allow the signal to be accurately reconstructed, and if
there is, how close is close enough. To get a better understanding of
undersampling these matters, we need to consider an alternative way of representing
a signal. Later, this will also help us to understand some related aspects
of sound and image processing.
You are probably familiar with the idea that a musical note played on
an instrument consists of waveforms of several different frequencies
added together. There is the fundamental, which is the pitch associated 18
40. Digitization
24 F U N DA M E N TA L S
¡กÒาÃร undersampl
the same level. The effects of such undersampling
ingon¢ขthe §งáแËหin which
Íอ way Åล‹‹§ง
¡กÓำàเ¹นÔิ´ด ¡ก‹‹Íอ perceived depend on
the reconstructed signal will be ãใËหŒŒàเ¡กÔิ´ด»ป˜˜ÞญËห what the signal
r– construct ÊสÑัÞญ
represents esound, image, and so on – and whether is
Òาitãใ¹นtime-varying
¡กÒาÃร
Þญ instances later. Suffice Åล
or space-varying. We will describe specific Òา³ณ·ท∙Õี่¨จÐะáแÊส´ด§ง it to
¼ผ
say, for now, that they are manifested as distortions and artefacts which
are always undesirable.
si si+1 It is easy enough to see that if the sampling rate is too low some detail
will be lost in the sampling. It is less easy to see whether there is ever
Figure 2.5. Undersampling
Íอ§ง output ¨จÐะ
any rate at which we can be sure that the samples are close enough
Ðะ¡กÑั¹น¤คØุ³ณÀภÒา¾พ¢ข
together to allow the signal to be accurately reconstructed, and if
àเ¾พ×ื่Íอ»ปÃร
undersampling ªชŒŒÍอÑัµตÃรÒา¡กÒาÃร sa mpling ´ดŒŒÇวÂย
there is, how close is close enough. To get a better understanding of
µตŒŒÍอ§งãใ
these matters, we need to consider an alternative way of representing
«ซÖึ่§งàเ·ท∙‹‹Òา¡กÑัºบ 2f
h
a signal. Later, this will also help us to understand some related aspects
Nyquist rate
of sound and image processing.
You are probably familiar with the idea that a musical note played on
an instrument consists of waveforms of several different frequencies
added together. There is the fundamental, which is the pitch associated 18
53. CHAPTER
2 D I G I TA L DATA
Figure 2.12. Posterization
between areas of those colours would be elided. The effect on black and white images can be
seen clearly in Figure 2.11, which shows a gradient swatch using 256, 128, 64, 32, 16, 8, 4, and
2 different grey levels. The original gradient varies linearly from pure white to pure black, and
as we reduce the number of different greys, you can see how values band together as they are
quantized increasingly coarsely. In a less regular image, the effect manifests itself as posterization, 24
56. Compression
You will learn, as we examine the individual media types in detail, that a characteristic property
Compression of media data is that it occupies a lot of storage. This means, in turn, that it needs a lot of band-
width when it is transferred over networks. Storage and bandwidth are limited resources, so the
high demands of media data can pose a problem. The common response to this problem is to
apply some form of compression, which means any operation that can be performed on data to
reduce the amount of storage required to represent it. If data has been compressed, an inverse
decompression operation will be required to restore it to a form in which it can be displayed or
used. Software that performs compression and decompression is often called a codec (short for
compressor/decompressor), especially in the context of video and audio.
Compression algorithms can be divided into two classes: lossless and
lossy. A lossless algorithm has the property that it is always possible to
original data decompress data that has been compressed and retrieve an exact copy
of the original data, as indicated in Figure 2.13. Any compression algo-
rithm that is not lossless is lossy, which means that some data has been
compress discarded in the compression process and cannot be restored, so that the
decompressed data is only an approximation to the original, as shown
in Figure 2.14. The discarded data will represent information that is
decompress
not significant, and lossy algorithms which are in common use do a
remarkable job at preserving the quality of images, video and sound,
compressed data
even though a considerable amount of data has been discarded. Lossless
Figure 2.13. Lossless compression algorithms are generally less effective than lossy ones, so for most multi-
media applications some lossy compression will be used. However, for
text the loss of even a single bit of information would be significant, so there is no such thing as
lossy text compression.
It may not be apparent that any sort of data can be compressed at all without loss. If no informa-
tion is being discarded, how can space be saved? In the case of text, lossless compression works 26
57. 2
Compression
CHAPTER DIGITAL DATA 31
You will learn, as we examine the individual media types in detail, that a characteristic property
Compression of media data is that it occupies a lot of storage. This means, in turn, that it needs a lot of band-
width when it is transferred over networks. Storage and bandwidth are limited resources, so the
with directly, so storage schemes which make less than optimal use of
the available bits are usedhigh demands of instead. data can pose a problem. The common response to this problem is to
most of the time media
original data
apply some form of compression, which means any operation that can be performed on data to
Lossless and lossy compression arethe amount separate techniques. Mostrepresent it. If data has been compressed, an inverse
reduce not entirely of storage required to
lossy algorithms make use of some lossless technique as part required to restore it to a form in which it can be displayed or
decompression operation will be of the total
compression process. Generally,Software that performs compression and decompression is often called a codec (short for
used. once insignificant information has been compress
discarded, the resulting data is more amenable to lossless compression. context of video and audio.
compressor/decompressor), especially in the
This is particularly true in the case of image compression, as we will
explain in Chapter 4. Compression algorithms can be divided into two classes: lossless and
lossy. A lossless algorithm hascompressed datathat it is always possible to
the property
Ideally, lossy compression will data be applied atdecompresspossible
original only the latest data that has been compressed and retrieve an exact copy
stage in the preparation of the media for delivery. Any processing that as indicated in Figure 2.13. Any compression algo-
of the original data, decompress
is required should be done on uncompressed or losslesslythat is not lossless is lossy, which means that some data has been
rithm compressed
data whenever possible. There are two reasons for this. When data is
compress discarded in the compression process and cannot be restored, so that the
lossily compressed the lost information can never be retrieved, which
decompressed data is only an approximation to the original, as shown
means that if data is repeatedly compressed and decompressed in this
in Figure 2.14. The discarded data will represent information that is
way its quality will gradually deteriorate. Additionally, some processing
decompress
not significant, and lossy algorithms which are in common use do a
operations can exaggerate the loss of quality caused by some types of
remarkable uncom- preserving the qualitydata images, video and sound,
compression. For both these reasons, it is best to work with
job at decompressed of
compressed data
pressed data, and only compress it for final delivery. even though a considerable amount of data has been discarded. Lossless
Figure 2.13. Lossless compression algorithms are generally less effective than lossy ones, so for most multi-
media applications Figure 2.14. Lossy compression
This ideal cannot always be achieved.Video data is usually compressed in some lossy compression will be used. However, for
the camera, and although text the loss of even a produce uncompressed
digital still cameras that single bit of information would be significant, so there is no such thing as
lossy text compression.
images are increasingly common, many cheaper cameras will compress photographs fairly severely
when the pictures are being taken. It may be necessary for a photographer to allow the camera to
compress images, in order to fit thembe apparent that any sort of data these circumstances, data without loss. If no informa-
It may not onto the available storage. Under can be compressed at all
should be decompressed tion isand an uncompressed version should be used In the case of text, lossless compression works
once, being discarded, how can space be saved? for working. This 26