CS 411 : Data Compression

Lecture 6

Scalar and Vector

Quantization
Lossy Image Compression Techniques

Scalar quantization (SQ)

Vector quantization (VQ)
Discrete Cosine Transform (DCT)
Compression:
 JPEG
Wavelet Compressions
 Quantization

 Definition:
 Quantization: a process of representing a large –
possibly infinite – set of values with a much smaller set.
 Scalar quantization: a mapping of an input value x into
a finite number of output values, y:
Q: x y

 One of the simplest and most general idea in

lossy compression.
 Scalar Quantization

Many of the fundamental ideas of

quantization and compression are easily
introduced in the simple context of scalar
quantization.
An example: any real number x can be
rounded off to the nearest integer, say
q(x) = round(x)
Maps the real line R (a continuous space)
into a discrete space.
 Quantizer

 The design of the quantizer has a significant impact

on the amount of compression obtained and loss
incurred in a lossy compression scheme.
 Quantizer: encoder mapping and decode mapping.
 Encoder mapping
 – The encoder divides the range of source into a
number of intervals
– Each interval is represented by a distinct codeword
 Decoder mapping
– For each received codeword, the decoder
generates a reconstruct value
 Components of Quantization

1. Encoder mapping: Divides the range of values

that the source generates into a number of
intervals. Each interval is then mapped to a
codeword. It is a many-to-one irreversible
mapping. The code word only identifies the
interval, not the original value. If the source or
sample value comes from a analog source, it is
called a A/D converter.
Mapping of a 3-bit Encoder

Codes

000 001 010 011 100 101 110 111

-3.0 -2.0 -1.0 0 1.0 2.0 3.0 input

Mapping of a 3-bit D/A Converter

Input Codes Output

000 -3.5
001 -2.5
010 -1.5
011 -0.5
100 0.5
101 1.5
110 2.5
111 3.5
 Components of a Quantizer

2. Decoder: Given the code word, the decoder

gives a an estimated value that the source might
have generated. Usually, it is the midpoint of the
interval but a more accurate estimate will
depend on the distribution of the values in the
interval. In estimating the value, the decoder
might generate some errors.
Quantization
 Why? -- To reduce number of bits per sample
F’(u,v) = round(F(u,v)/q(u,v))
 Example: 101101 = 45 (6 bits).
Truncate to 4 bits: 1011 = 11. (Compare 11 x 4 =44 against 45)
Truncate to 3 bits: 101 = 5. (Compare 8 x 5 =40 against 45)
Note, that the more bits we truncate the more precision we lose
 Quantization error is the main source of the Lossy Compression.
 Uniform Quantization:
 q(u,v) is a constant.
 Non-uniform Quantization -- Quantization Tables
 Eye is most sensitive to low frequencies (upper left corner in
frequency matrix), less sensitive to high frequencies (lower right
corner)
 Custom quantization tables can be put in image/scan header.
 JPEG Standard defines two default quantization tables, one each for
luminance and chrominance.
10
Scalar Quantization

 Problem 1:
You’re given 16-bit integers (0-65545). Unfortunately,
you only have space to store 8-bit integers (0-255).
Come up with a representation of those 16-bit
integers that uses only 8 bits!
 Problem 2:
You have a string of those 8-bit integers that use your
representation. Recreate the 16-bit integers as best
you can!
Uniform Quantization
Uniform Quantizer

Zero is one of the output levels Zero is not one of the output levels
M is odd M is even
Scalar Quantization

 Build a codebook with a training set, then

always encode and decode with that fixed
codebook.
 Build a codebook for each image and
transmit the codebook with the image.
Uniform Scalar Quantization Example
Uniform Scalar Quantization Example
Uniform Scalar Quantization Example 1
Non-Uniform Scalar Quantization Example 2
Image Compression

Original 8bits/pixel 3bits/pixel

Image Compression

2bits/pixel 1bit/pixel
 Vector Quantization

 The data stream is divided to blocks called vectors.

 Table, called code-book
contains a set of patterns
may be predefined or dynamically constructed.
 Find best matching pattern in the table.
 Send table entry number instead of vector.
Principle of Vector Quantization
Original Code-book Compressed
data data stream
stream Pattern 0
Vector 0 of data stream k
...
Vector 1 of data stream 0
Pattern j
...
... ...
...
... Pattern k
i
... ...
Vector n of data stream ...
Pattern P
Vector Quantization with Error Transmission

Original Code-book Compressed

data data stream
stream
Pattern 0
k Er0
Vector 0 of data stream ...
0 Er1
Vector 1 of data stream Pattern j ... ...
... ... ... ...
... Pattern k i Eri
... ...
Vector n of data stream ...
Pattern P
Vector Quantization

Quantization is a mapping of a large set of

values to a smaller set of values.
Vector Quantization
Vector Quantization
Vector Quantization
Vector Quantization
Vector Quantization
Compression Formula

 Amount of compression:
 Codebook size is K, input vector of dimension L
 Inorder to inform the decoder of which code
vector is selected, we need to use log 2 K  bits.
 E.g. need 8 bits to represent 256 code vectors.
 Rate: each code vector contains the log 2 K  / L
reconstruction value of L source output samples,
the number of bits per sample would be:
.
 Sample: a scalar value in vector.
 K: level of vector quantizer.
 Vector Quantization Example

 The original image has 256 gray scale is

600*600 pixels. The image is divided into
blocks each of size 4*4 pixels. The codebook
which will be used to reconstruct the image
consists of 32 vectors (32 blocks each of size
4*4)
 Vector Quantization Example
 Labels size= number of labels * number of bits per label= 22500 *
5=112500 bits = 14063 Bytes
 Codebook storage size = number of vectors * vector size *
number of bits per pixel= 32*(4*4)*8=4096 bits = 512 bytes
 Total compressed image size= label size + code book storage
size= 112500+4096=116596 nbits = 14575 bytes
 Original image size = 600*600*8 = 2880000 bits = 36000bytes
 Compression ratio = original/compressed = 36000 / 14575= 24.7 :
1

 When label size is much greater than codebook storage size , codebook
storage size can be neglected during compression ratio calculations
 Note: codebook storage size is independent on image size , number of
labels depends on image size
 Labels size= number of labels * number of bits per label= 22500 *
5=112500 bits = 14063 Bytes
 Codebook storage size = number of vectors * vector size * number of bits
per pixel= 32*(4*4)*8=4096 bits = 512 bytes
 Total compressed image size= label size = 112500 bits = 14063 bytes
 Original image size = 600*600*8 = 2880000 bits = 36000bytes
 Compression ratio =original/compressed=360000/14063= 25.59 :
1