By

Prof (Dr.) P P Ghadekar

 OUTLINE
 On completion Student will be able to understand
 Basic concepts of DCT
 Different types of DCT
 JPEG Coding
 Application of DCT

2 Prof P P Ghadekar
 Transform Characteristic
 Transforms are useful entities that encapsulate these characteristics
 Data decorrelation
 Data-independent basis functions
 Fast implementation
 Linearity
 Orthogonality

3 Prof P P Ghadekar
 Data decorrelation
 The ideal transform completely decorrelates the data in a
sequence/block; i.e., it packs the most amount of energy in the
fewest number of coefficients.
 In this way, many coefficients can be discarded after quantization
and prior to encoding.
 It is important to note that the transform operation itself does not
achieve any compression.
 It aims at decorrelating the original data and compacting a large
fraction of the signal energy into relatively few transform
coefficients.

4 Prof P P Ghadekar
 Fast Implementation
 The number of operations required for an n-point transform
is generally of the order O(n2).(order of n2 time complexity)
 Some transforms have fast implementations, which reduce
the number of operations to O(n log n).
 For a separable n × n 2-D transform, performing the row
and column 1-D transforms successively reduces the
number of operations from O(n4) to O(2n2 log n).

O-Big O notation used to represent efficiency (No of steps)

6 Prof P P Ghadekar
 Linearity and Orthogonality
 Linearity
 Linearity principle allows one to one mapping between
pixel values and transform coefficients.

 Orthogonality
 Orthogonal transform have the feature of eliminating
redundancy in the transformed image.

7 Prof P P Ghadekar
 Transform based Image Coding Scheme
Original
Image TTransformation
Quantiser Entropy
Enencoder Channel
m Encoder

Source Encoder

Transmission
Channel
Reconstructed
Image

Inverse Inverse EntropyEn

decoder Channel
Transformationm Quantiser Decoder

8 Source Decoder
Prof P P Ghadekar
 Transform based Image Coding Scheme
 Quantisation
 It is the process of reducing the number of possible values of a
quantity thereby reducing the number of bits needed to represent.
 It is an irreversible process.
 Entropy Encoder
 The purpose of an entropy encoder is to reduce the number of
bits required to represent each symbol at the quantiser output.
 Commonly used entropy coding techniques are Huffman coding,
Arithmetic coding, run length coding etc.
 It is a lossless coding scheme.

9 Prof P P Ghadekar
 Introduction
 A Discrete cosine transform (DCT) expresses a sequence
of finitely many data points in terms of a sum of cosine
functions oscillating at different frequencies.
 The discrete cosine transforms (DCT) and discrete sine
transform (DST) are members of a family of sinusoidal
unitary transforms.
 They are real, orthogonal, and separable with fast algorithms
for its computation.
 They have a great relevance to data compression.
 The discrete cosine transform (DCT) is a Fourier-related
transform similar to the discrete Fourier transform (DFT),
but using only real numbers.
10 Prof P P Ghadekar
 Introduction
 It is equivalent to a DFT of roughly twice the length,
operating on real data with even symmetry.
 There are eight standard variants, of which four are
common.

11 Prof P P Ghadekar
 WHY TO USE DCT?
 We use DCT to decorrelate the input signal efficiently.

12 Prof P P Ghadekar
 DFT Basis Images

Real (COS) Imaginary (SIN)

13 Prof P P Ghadekar
 DCT Basis Images

14 Prof P P Ghadekar
 Types of DCT
 The family of discrete trigonometric transforms consists of 8 versions of DCT.
 Each transform is identified as EVEN or ODD and of type I, II, III, and IV.
 All present digital signal and image processing applications (mainly transform
coding and digital filtering of signals) involve only even types of the DCT and
DST.
 Therefore, we consider these four even types of DCT.

15 Prof P P Ghadekar
 Inverse transforms:
 The inverse of DCT-I is DCT-I multiplied by 2/(n-1).
 The inverse of DCT-IV is DCT-IV multiplied by 2/n. The
inverse of DCT-II is DCT-III multiplied by 2/n (and vice
versa).
 Like for the DFT, the normalization factor in front of
these transform definitions is merely a convention and
differs between treatments.
 For example, some authors multiply the transforms by
√(2/n) so that the inverse does not require any additional
multiplicative factor.

16 Prof P P Ghadekar
 Why DCT-II is Common?
 The DCT, and in particular the DCT-II, is often used in
signal and image processing, especially for lossy data
compression.
 Because it has a strong "energy compaction" property.
 Most of the signal information tends to be concentrated in
a few low-frequency components of the DCT, approaching
the Karhunen-Loève transform.

17 Prof P P Ghadekar
 Why DCT-II is Common?
 Performance of DCT-II is closest to the statistically optimal KLT
based on a number of performance criteria.
o Variance distribution,
o Energy packing efficiency,
o Residual correlation,
o Rate distortion,
o maximum reducible bits …
 Exhibition of desirable characteristics for data compression
namely,
o Data decorrelation
o Data-independent basis functions
o Fast implementation

18 Prof P P Ghadekar
 Why DCT-II is Common?
 The importance of DCT II is further highlighted by its -
o Superiority in bandwidth compression (redundancy
reduction) of a wide range of signals.
o Powerful performance in the bit-rate reduction.
o Existence of fast algorithms for its implementation.
 DCT-II and its inversion, DCT-III, have been employed in
the international image/video coding standards: e.g.: JPEG,
MPEG, H.261, H.263, H.264…

19 Prof P P Ghadekar
 1-D DCT (DCT-II)

1-D DCT
2 N 1
(2i  1)k
X (k )  C (k ) x(i ) cos[ ]
N i 0 2N

1-D IDCT
N 1
(2i  1)k

2
x(i )  C (k ) X (k ) cos[ ]
N k 0
2N

k = 0, 1, 2, …, N-1.
and i = 0, 1, 2, …, N-1.

20 Prof P P Ghadekar
 Orthogonality
C is orthogonal which Implies C-1 = CT and that
Entails:
CTC = CCT = I
This property is used to solve matrix equations easily.

21 Prof P P Ghadekar
 1-D DCT
The discrete cosine transform, C, has one basic characteristic: it
is a real orthogonal matrix.

 1 1 1 
  
2 2 2
 (2 n  1) π 
2  cos π cos
3π
 cos
C
n 2n 2n 2n 
     
 (n  1) π (n  1)3 π (n  1)(2 n  1) π 
 cos cos  cos 
2n 2n 2n

 1  (n  1) 
 cos  cos 
2 2n 2n
 1 3 (n  1)3 
2 cos  cos
C 1  C T 
n  2 2n 2n 

   
 1 (2n  1) (n  1)(2n  1) 
 cos  cos 
 2 2n 2n 

22 Prof P P Ghadekar
 Interpolation with DCT

 What is Interpolation?
Interpolation is a method of constructing new data points
within the range of a discrete set of known data points.

 Why interpolate with DCT?

DCT interpolation gives terms already arranged in terms of
importance to the human visual system !!
First terms are most important, last terms are least important.

23 Prof P P Ghadekar
 Discrete Cosine Transform (DCT)
The most important tool for overcoming drawbacks of
the 2D DFT is the 2D DCT.
There are several definitions of this transform in practice
and here we will use the following (for the 1D DCT):

24 Prof P P Ghadekar
 Inverse DCT
 Inverse DCT is defined as:

 DCT and inverse DCT are mutually inverse.

25 Prof P P Ghadekar
 2D DCT
 There is not unique form of the 2D DCT. The simplest
realization technique is calculation of the 1D DCT
along rows and after that along columns of newly
obtained matrix.
 However, there are alternative techniques for direct
evaluation of the 2D DCT. Again there are several
definitions of the 2D DCT that can be used in practice
but here we adopted:

26 Prof P P Ghadekar
 Inverse 2D DCT

 Inverse 2D DCT (for our the 2D DCT form) is:

27 Prof P P Ghadekar
 Mathematical Properties-DCT
 DCT Matrices are real and orthogonal.

 Unitary Property A-1 = A*T A* = Conjugate of A

 Linearity Property
M(αg + βf) = α Mg + βMf
For a matrix M, constants α and β, and vectors g and f, all DCTs are linear
transforms.
 The Convolution-Multiplication Property
Convolution in the spatial domain is equivalent to taking an inverse
transform of the product of forward transforms of two data sequences.
The convolution — multiplication property is a powerful tool for
performing digital filtering in the transform domain.

28 Prof P P Ghadekar
 Mathematical Properties-DCT
 All DCTs are separable transforms.-Multidimensional transform
can be decomposed into successive application of one-
dimensional (1-D) transforms in the appropriate directions.
 It is separable meaning that it can be separated into a pair of 1-D
DCTs.
 To obtain the 2-D DCT of a block a 1-D DCT is first performed
on the rows of the block then a 1-D DCT is performed on the
columns of the resulting block.
 The same applies to the IDCT.
 This process is illustrated on the following slide.

29 Prof P P Ghadekar
30 Prof P P Ghadekar
 Columns (256 pixels)
Raw Block of Pixels
128 127 130 128 134 130 128 128

128 127 130 128 134 130 128 128

An 8x8
125 126 130 127 130 131 128 127
block
Rows (256 pixels)

124 128 127 126 129 130 127 127

131 132 130 129 132 128 129 131

pixel[n] 129 131 134 131 133 129 131 131

129 134 134 136 136 137 134 132

pixel
[256+n] 133 135 136 138 137 140 135 132

pixel[n+1]

Part of a picture

Final result after DCT by columns Result after DCT by rows

1024 -4 -11 0 -4 -6 0 1 365 -1 -4 1 1 -3 1 3

-18 -1 1 0 2 -2 2 4 365 -1 -4 1 1 -3 1 3

11 -1 -5 2 0 0 1 -1 362 -3 -4 0 -3 -1 2 1

0 2 2 -3 3 -2-1 -2 3 359 -3 -3 0 -3 -3 0 -1

0 1 1 1 0 -1 -1 -2 368 1 1 -1 1 -3 -2 1

-4 -3 -3 -1 -3 -1 0 -2 370 0 -3 -3 -1 -1 -1 2

-2 0 -1 -2 -1 1 1 0 378 -3 -6 0 -3 -2 -1 -2

2 0 1 0 0 -1 -1 -1 383 -1 -6 2 -3 0 0 -3

31 Prof P P Ghadekar
 DCT
 For example, the DCT is used in JPEG image
compression, MJPEG video compression, and MPEG
video compression.
 There, the two-dimensional DCT-II of 8x8 blocks is
computed and the results are quantized and entropy coded.
 In this case, n is typically 8 and the DCT-II formula is
applied to each row and column of the block.
 The result is an 8x8 transform coefficient array in which
the (0,0) element is the DC (zero-frequency) component
and entries with increasing vertical and horizontal index
values represent higher vertical and horizontal spatial
frequencies.

32 Prof P P Ghadekar
 DCT– Matrix Form
DCT-Matrix Form

IDCT-Matrix Form

33 Prof P P Ghadekar
 DCT Result

Prof P P Ghadekar 34
 Example of a 4x4 DCT Matrix
Example of a 4x4 DCT Matrix

Example of a 4x4 IDCT Matrix

35 Prof P P Ghadekar
 Example of a 4x4 DCT Matrix
 N=4 point DCT can be generated by
c[n,m]= a[n] cos((2m+1)nП /8)

Assume the signal is X=[0,1,2,3]T, then its DCT transform is

DCT Matrix Data Transform

36 Prof P P Ghadekar Vector Vector
 Example of a 4x4 DCT Matrix
 The inverse transform is-

X=

DCT Matrix Transform Data

Vector Vector

37 Prof P P Ghadekar
 Joint Photographic Expert Group

JPEG involves the following four distinct steps

 Block Preparation
 Discrete Cosine Transform
 Quantization
 Compression

Input Block Discrete Cosine Data Output

Quantization
Preparation Transform Compression

38 Prof P P Ghadekar
 JPEG Coding
To perform the JPEG coding, an image (in colour or
gray scales) is first subdivided into blocks of 8x8
pixels.
The Discrete Cosine Transform (DCT) is then
performed on each block.
This generates 64 coefficients which are then
quantized to reduce their magnitude.

39 Prof P P Ghadekar
 JPEG Coding
The coefficients are then reordered into a one-
dimensional array in a zigzag manner before further
entropy encoding.
The compression is achieved in two stages; the first is
during quantization and the second during the entropy
coding process.
JPEG decoding is the reverse process of coding.

40 Prof P P Ghadekar
 JPEG Coding Block Diagram

41 Prof P P Ghadekar
 JPEG-Block Preparation
 Reasons for Block Processing
 Taking DCT for an entire image requires large memory.
 Taking DCT for an entire image is not a good idea for
compression due to spatially varying statistics within an image.

 Advantages of Block Processing

 The transform of small blocks is much easier to compute than
the complete image.
 The pixel correlation will not exceed usually 16 or 32 pixels.
 Correlation between neighbouring pixels is more than distinct
pixels.

42 Prof P P Ghadekar
 JPEG-Block Preparation

Luminance (Brightness)
Y=0.30R + 0.59G + 0.11B
Chrominance (Colour) PAL
U= -0.18R - 0.29G +0.44B
V=0.62R - 0.52G - 0.10B

43 Prof P P Ghadekar
 JPEG-DCT
Each block of 64 pixels goes through a transformation called DCT.
To understand the nature of this transformation, let us consider the result of the
transformations for three different cases-
1)Uniform Grayscale
2)Two sections
3)Gradient Grayscale.

Case-I Uniform Grayscale

44 Prof P P Ghadekar
 JPEG-DCT
Case-II Two Section

Case-III Gradient Grayscale

45 Prof P P Ghadekar
 Quantization

 After the T table is created, the values are quantized to reduce the
number of bits needed for encoding.
 Quantization divides the number of bits by a constant and then
drops the fraction. This reduces the required number of bits even
more.
 In most implementations, a quantizing table (8 by 8) defines how
to quantize each value.
 The divisor depends on the position of the value in the T table.
 This is done to optimize the number of bits and the number of 0s
for each particular application.

46 Prof P P Ghadekar
 JPEG-Quantization
Quantization

47 Prof P P Ghadekar
 Compression

 After quantization the values are read from the table, and
redundant 0s are removed.
 However, to cluster the 0s together, the process reads the table
diagonally in a zigzag fashion rather than row by row or column
by column.
 The reason is that if the picture does not have fine changes, the
bottom right corner of the T table is all 0s.
 JPEG usually uses run-length encoding at the compression phase
to compress the bit pattern resulting from the zigzag linearization.

48 Prof P P Ghadekar
 Run Length Coding Result-20 1515 12 17 12 A 0 58
49 Prof P P Ghadekar
 JPEG-Run length Encoding
A Zigzag scanning pattern is used to concentrate all the 0’s together. The runs of
0’s can be replaced by a single count (say,38 0’s).

Zigzag scanning

Run Length Coding

160 95 90 28 85 30 6 20 24 4 1 3 16 4 1 0 1 3 3 A 0 5 2 1 A 0 38
Another Example-
50 21000000000000000000000000003 21A0263
51 Prof P P Ghadekar
 Bit Allocation
 There are two methods used for bits allocation
 Zonal coding
 Threshold coding

52 Prof P P Ghadekar
 Zonal coding
 Zonal coding
 It is based on the fact that the transform coefficients of
maximum variance carry the most picture information.
 The locations of the coefficients with the largest variances are
indicated by means of a zonal mask, which is same for all
blocks.
 All transform coefficients in the zone are retained, while all
coefficient out side the zone are set to zero.
 To design a zonal mask, variances of each coefficient can be
calculated based on a global image model, such as the Guass-
Markov model.

53 Prof P P Ghadekar
 Bit Allocation in Zonal Coding
 A simple bit allocation strategy is to choose the number of bits
proportional to the variance of each coefficient over the blocks.
 If the number of retained coefficients are M with the variances
σi2 , i=1,2,3…….M. Then the number of bits allocated for each of
these coefficients is given by

Where B is the total number of bits available to represent a block

54 Prof P P Ghadekar
 Zonal coding
Information theory says coefficients with maximum variance carry
the most information.
15 out of 64 transform coefficients , with largest variance, are kept

keep

1 1 1 1 1 0 0 0
1 1 1 1 0 0 0 0
1 1 1 0 0 0 0 0
1 1 0 0 0 0 0 0 discard
1 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0
Prof P P Ghadekar
0 0 0 0 0
55
 Quantization
 Retained coefficients are quantized and coded
 Coefficients are either normalized by their standard
deviation and uniformly quantized.

56 Prof P P Ghadekar
 Bit allocation
 How many bits should be assigned to each coefficient?
 Information theory tells us, that a Gaussian random variable,
subject to distortion D, cannot be represented by fewer bits.
 Large variance coefficients need more bits. Small Variance
coefficients need less bits.

1  2 
log 2   bits
2  D 

57 Prof P P Ghadekar
 Bit allocation table

8 7 6 4 3 2 1 0
7 6 5 4 3 2 1 0
6 5 4 3 2 1 1 0
4
3
2
1
0

Number of bits per coefficient

58 Prof P P Ghadekar
 Problem with zonal coding
 If we consider a fixed area we run the risk of cutting off
occasional transform coefficients that are large enough to
be included

excluded

59 Prof P P Ghadekar
 Threshold coding
 One solution is to threshold coefficients and possibly use
different thresholds for different subimages.
 This way we will end up keeping those coefficients that
exceed threshold, wherever they might be.

60 Prof P P Ghadekar
 Threshold Coding
 In most images, different blocks have different spectral and
statistical characteristic.
 So it is necessary to use adaptive bit allocation methods.
 In threshold coding, each transform coefficient is compared
with a threshold.
 If it is smaller than the threshold then it is set to zero.
 If it is larger than the threshold, it will be retained for
quantization and encoding.
 The thresholding method is an adaptive method where only
those coefficient whose magnitude are above a threshold are
retained within each block.
61 Prof P P Ghadekar
 Quantiser
 The purpose of quantisation is to remove the components of the
transformed data that are unimportant to the visual appearance of
the image and to retain the visually important components.
 The quantiser utilizes the fact that the human eye is unable to
perceive some visual information in an image.
 Such information is deemed redundant and can be discarded
without noticeable visual artefacts.
 Such redundancy is referred to as psychovisual redundancy.
 The process of assigning a particular sample to a particular level is
called quantisation.
 It is non-linear and irreversible. It is a lossy compression Scheme.
 Rounding the number to the nearest integer can be represented
with minimum number of bits
62 Prof P P Ghadekar
 Picking the N-largest transform coefficients
 In a 2D array, how do we pick N-largest coefficients.
 Answer: zig-zag scan. Coefficients are rearranged in a 1D-
sequence

0 1 5 6 14 15 27 28
2 4 7 13 16 26 29
3 8 12 17 25 30
9 11 18 24 31
10 19 23 32
20 22 33
21 34
35

63 Prof P P Ghadekar
 Threshold details
 How do we threshold the transform coefficients?
 There are 3 ways
 Single global threshold for all subimages
 Different thresholds for different subimages
 Different threshold for different locations in the subimage

64 Prof P P Ghadekar
 Image Compression
 Image compression is a method that reduces the amount of
memory it takes to store in image.

 We will exploit the fact that the DCT matrix is based on our
visual system for the purpose of image compression.

 This means we can delete the least significant values without

our eyes noticing the difference.

67 Prof P P Ghadekar
 Image Compression
 Now we have found the matrix Y = C(CXT)T

 Using the DCT, the entries in Y will be organized based on the

human visual system.

 The most important values to

our eyes will be placed in the Most
upper left corner of the matrix.
Important
Semi-
 The least important values
will be mostly in the lower Important

right corner of the matrix. Least

Important

68 Prof P P Ghadekar
 Image Compression

8 x 8 Pixels Image

69 Prof P P Ghadekar
 Image Compression
 Gray-Scale Example
 Value Range 0 (black) --- 255 (white)

63 33 36 28 63 81 86 98
27 18 17 11 22 48 104 108
72 52 28 15 17 16 47 77
132 100 56 19 10 9 21 55
187 186 166 88 13 34 43 51
184 203 199 177 82 44 97 73
211 214 208 198 134 52 78 83
211 210 203 191 133 79 74 86

X
70 Prof P P Ghadekar
 Image Compression
 2D-DCT of matrix
Numbers are coefficients of
polynomial

-304 210 104 -69 10 20 -12 7

-327 -260 67 70 -10 -15 21 8
93 -84 -66 16 24 -2 -5 9
89 33 -19 -20 -26 21 -3 0
-9 42 18 27 -7 -17 29 -7
-5 15 -10 17 32 -15 -4 7
10 3 -12 -1 2 3 -2 -3
12 30 0 -3 -3 -6 12 -1

Y
71 Prof P P Ghadekar
 Image Compression
 Cut the least significant components

-304 210 104 -69 10 20 -12 0

-327 -260 67 70 -10 -15 0 0
93 -84 -66 16 24 0 0 0
89 33 -19 -20 0 0 0 0
-9 42 18 0 0 0 0 0
-5 15 0 0 0 0 0 0
10 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0

As you can see, we save a little over half the original memory.

72 Prof P P Ghadekar
 Reconstructing the Image
 New Matrix and Compressed Image

55 41 27 39 56 69 92 106
35 22 7 16 35 59 88 101
65 49 21 5 6 28 62 73
130 114 75 28 -7 -1 33 46
180 175 148 95 33 16 45 59
200 206 203 165 92 55 71 82
205 207 214 193 121 70 75 83
214 205 209 196 129 75 78 85

73 Prof P P Ghadekar
 Can You Tell the Difference?
Original Compressed

74 Prof P P Ghadekar
 Image Compression

Original Compressed

75 Prof P P Ghadekar
 Linear Quantization
 We will not zero the bottom half of the matrix.
 The idea is to assign fewer bits of memory to store
information in the lower right corner of the DCT matrix.

76 Prof P P Ghadekar
 Linear Quantization

Use Quantization Matrix (Q)

Q=p* 8 16 24 32 40 48 56 64
16 24 32 40 48 56 64 72
24 32 40 48 56 64 72 80
32 40 48 56 64 72 80 88
40 48 56 64 72 80 88 96
48 56 64 72 80 88 96 104
56 64 72 80 88 95 104 112
64 72 80 88 96 104 112 120

77 Prof P P Ghadekar
 Linear Quantization
p is called the loss parameter

It acts like a “knob” to control compression

The greater p is the more you compress the

image

78 Prof P P Ghadekar
 Linear Quantization
We divide the each entry in the DCT matrix by the
Quantization Matrix

-304 210 104 -69 10 20 -12 7 8 16 24 32 40 48 56 64

-327 -260 67 70 -10 -15 21 8 16 24 32 40 48 56 64 72
93 -84 -66 16 24 -2 -5 9 24 32 40 48 56 64 72 80
32 40 48 56 64 72 80 88
89 33 -19 -20 -26 21 -3 0
40 48 56 64 72 80 88 96
-9 42 18 27 -7 -17 29 -7 48 56 64 72 80 88 96 104
-5 15 -10 17 32 -15 -4 7 56 64 72 80 88 95 104 112
10 3 -12 -1 2 3 -2 -3 64 72 80 88 96 104 112 120
12 30 0 -3 -3 -6 12 -1

79 Prof P P Ghadekar
 Linear Quantization
p=1 p=4

-38 13 4 -2 0 0 0 0 -9 3 1 -1 0 0 0 0
-20 -11 2 2 0 0 0 0
-5 -3 1 0 0 0 0 0
1 -1 0 0 0 0 0 0
4 -3 -2 0 0 0 0 0 1 0 0 0 0 0 0 0
3 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0

New Y: 14 terms New Y: 10 terms

80 Prof P P Ghadekar
 Linear Quantization
p=1 p=4

81 Prof P P Ghadekar
 Linear Quantization
p=1 p=4

82 Prof P P Ghadekar
 Properties of DCT
 Decorrelation - The principle advantage of image
transformation is the removal of redundancy between
neighboring pixels. This leads to uncorrelated transform
coefficients which can be encoded independently.
 Energy Compaction - DCT exhibits excellent energy
compaction for highly correlated images. The uncorrelated
image has its energy spread out, whereas the energy of the
correlated image is packed into the low frequency region.

83 Prof P P Ghadekar
  Orthogonality - IDCT basis functions are orthogonal .
Thus, the inverse transformation matrix of A is equal to its
transpose i.e. invA= A'.
 Separability – Perform DCT operation in any of the
direction first and then apply on second direction,
coefficient will not change.

84 Prof P P Ghadekar
 Applications
 The DCT is used in JPEG image compression, MJPEG
video compression, and MPEG video compression.
 A related transform, the modified discrete cosine
transform, or MDCT, is used in MP3 audio compression.
 DCTs are also widely employed in solving partial
differential equations by spectral methods, where the
different variants of the DCT correspond to slightly
different even/odd boundary conditions at the two ends of
the array.

85 Prof P P Ghadekar
 Advantages of DCT
 The DCT does a better job of concentrating energy into
lower order coefficients than does the DFT for image data.
 The DCT is purely real, the DFT is complex.
 Assuming a periodic input, the magnitude of the DFT
coefficients is spatially invariant. This is not true for the
DCT.

86 Prof P P Ghadekar
 Compared with DFT, DCT has two main advantages
 It’s a real transform with better computational efficiency
than DFT which by definition is a complex transform.
 It does not introduce discontinuity while imposing
periodicity in the time signal.
 In DFT, as the time signal is truncated and assumed
periodic, discontinuity is introduced in time domain and
some corresponding artifacts is introduced in frequency
domain.
 But as even symmetry is assumed while truncating the time
signal, no discontinuity and related artifacts are introduced
in DCT.

87 Prof P P Ghadekar
 Drawbacks of DCT
 Blocking artefacts
 Graininess
 Blurring
Blocking Artefact mean there is variations in overall
characteristics from one block to another block i.e. one block
is having different DC level and another block is also having
DC level. This gives rise to some checkerboard effect.
Graininess-It causes due to truncation of low spectral
coefficient or lack of details.
Blurring effect-It is due to truncation of high frequency
components.

88 Prof P P Ghadekar
 JPEG 2000-Block Diagram

89 Prof P P Ghadekar
 Thanks

Prof (Dr) P P Ghadekar, VIT Pune

90