The DCT is conceptually similar to the DFT, except:

• 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 (magnitude and

phase).

• A DCT operation on a block of pixels produces coefficients that

are similar to the frequency domain coefficients produced by a
DFT operation.

• An N-point DCT has the same frequency resolution as and is

closely related to a 2N-point DFT. The N frequencies of a 2N
point DFT correspond to N points on the upper half of the unit
circle in the complex frequency plane.

• Assuming a periodic input, the magnitude of the DFT

Relationship between DCT and FFT
coefficients is spatially invariant (phase of the input does not
matter). This is not true for the DCT.
• DCT (Discrete Cosine Transform) is
actually a cut-down version of the FFT:
• For most images, after transformation the majority of
signal energy is carried by just a few of the low order • Only the real part of FFT
DCT coefficients. These coefficients can be more • Computationally simpler than FFT
finely quantized than the higher order coefficients.
Many higher order coefficients may be quantized to 0 • DCT -- Effective for Multimedia
(this allows for very efficient run level coding) Compression
• DCT MUCH more commonly used.

DCT Encoding The basic operation of the DCT is as follows:

The general equation for a 1D (N data items) DCT is
defined by the following equation: • The input image is N by M;
• f(i,j) is the intensity of the pixel in row i and column j;
N −1
⎛ π k (2n + 1) ⎞ • F(u,v) is the DCT coefficient in row k1 and column k2 of
X DCT [ k ] = ∑ 2 x[n] cos ⎜ ⎟, 0 ≤ k ≤ N − 1. the DCT matrix.
n =0 ⎝ 2N ⎠
• For most images, much of the signal energy lies at low
frequencies; these appear in the upper left corner of the
DCT.
• and the corresponding inverse 1D DCT transform is
• Compression is achieved since the lower right values
simple F-1(u), i.e.: represent higher frequencies, and are often small - small
enough to be neglected with little visible distortion.
1 N −1
⎛ π k (2n + 1) ⎞
x[ n] =
N
∑ α [k ] X
n =0
DCT [k ] cos ⎜
⎝ 2N
⎟,
⎠
0 ≤ n ≤ N − 1. • The DCT input is an 8 by 8 array of integers. This array
contains each pixel's gray scale level;
• 8 bit pixels have levels from 0 to 255.

1
It is computationally easier to implement and more
efficient to regard the DCT as a set of basis functions
which given a known input array size (8 x 8) can be
precomputed and stored.
This involves simply
computing values for a
convolution mask (8 x8
window) that get applied.
The values as simply
calculated from the DCT
formula. The 64 (8 x 8)
DCT basis functions are
illustrated in Fig.
• The N- point DFT X [k ] of a length-N real sequence is a
complex sequence satisfying the symmetry condition
X [k ] = X *[( k ) N ].
• For N even, the DFT samples X(0) and X[(N-2)/2] are real
and distinct. The remaining N-2 DFT samples are complex,
and only half of these samples are distinct as the rest are
complex conjugates of these samples.
• For N odd, the DFT sample X[0] is real, and the remaining
N-1 DFT samples are complex, of which only half of these
samples are distinct as the rest are complex conjugates of
these samples.
• As a result, there is a redundancy in the DFT-based
frequency-domain representation of a discrete time sequence.
• The transforms developed via this approach are called
Discrete Cosine Transform, often abbreviated as DCT and
discrete sine transforms often abbreviated as DST.
• There are four types of finite length sequences exhibiting
geometric symmetry, depending on the location of the
point of symmetry or antisymmetry :
• 1) whole sample symmetry (WS)
2) whole sample antisymmetry (WA)
3) half-sample symmetry (HS)
4) Half sample antisymmetry (HA)
Why DCT not FFT?
• DCT is similar to the Fast Fourier Transform (FFT), but
can approximate lines well with fewer coefficients.
• It is of interest to have orthogonal transform that represent
a real time-domain sequence X[k].
• Several such orthogonal transforms have been developed
with each having certain attractive properties.
Definition
• DFT of a real symmetric or antisymmetric finite length
sequence is a product of a linear-phase term and a real
amplitude function.
• One class of real orthogonal transforms is based on
converting an arbitrary sequence into either a symmetric or
an antisymmetric sequence and then extracting the real
orthogonal transform coefficients from the DFT of the
generated sequence with geometric symmetry.
• To develop a symmetric or antisymmetric sequence by
periodic extension from a specified finite length sequence,
the above type of symmetry or antisymmetry can be
applied at each end of the given sequence.
• This results in 16 distinct types of periodic extensions, of
which 8 extensions will be symmetric-periodic leading to 8
different types of DCTs, and 8 extensions will be
antisymmetric-periodic, leading to 8 different types of
DSTs.
• Here we consider only the symmetric periodic extensions.
2
 x[n]

• To illustrate the periodic extension process, xWSWS [n] n xHSHS [n]

consider a length-4 sequence x[n], shown in

figure below.

• The 8 symmetric periodic extensions xWSWA [n] x HSHA [n]

generated from {x[n]} are also shown in

figure.

Four symmetric periodic extensions of x[n].

(a) Periodic extension for DCT-1 (b) Periodic extension for DCT-2
(c) Periodic extension for DCT-3 (d) Periodic extension for DCT-4

xWSHS [n]
xHSWS [n]
• To develop the expression for the Discrete Cosine
Transform, for each of the above symmetric periodic
sequences, we first extract one period from each.

• For example to develop the Type-I Discrete Cosine

Transform, we extract one period y[n] of the symmetric
xWSHA [n]
xHSWA [n] periodic sequence xwsws [ n]. Thus if

{x[n]} = {a b c d},

then, the extracted period is given by,

Four anti-symmetric periodic extensions of x[n]. {y[n]} = {a b c d c b}.

(a) Periodic extension for DCT-5 (b) Periodic extension for DCT-6
(c) Periodic extension for DCT-7 (d) Periodic extension for DCT-8

• Likewise to develop the Type-2 Discrete Cosine Transform, • Let x[n] be a length-N sequence defined for 0 ≤ n ≤ N − 1.
we extract one period y[n] of the symmetric periodic • First x[n] is extended to a length 2N sequence by zero
sequence padding:
x HSHS [n].
⎧ x[ n], 0 ≤ n ≤ N −1
• For the length-4 sequence given above, the extracted period is xe [n] = ⎨
given by, ⎩0 N ≤ n ≤ 2 N − 1.
{y[n]} = {a b c d d c b a}
• Next a Type-2 symmetric sequence y[n] of length 2N is
• The DCT is then determined from the DFT of y[n]. formed from xe [n] according to
y[n] = xe [ n] + xe [2 N − 1 − n]. 0 ≤ n ≤ 2 N − 1.
• As the Type-2 DCT has been employed in a in a number of ⎧ x[n], 0 ≤ n ≤ N − 1,
international standards for image and video compression, such =⎨
as JPEG, MPEG and H.261, because of its better energy ⎩ x[2n − 1 − n]. N ≤ n ≤ 2 N − 1. .....(1)
compaction property, we consider here only this type of DCT,
Also called the even symmetrical discrete cosine transform.

3
• Note from (1), that the generated sequence y[n] satisfies • By making a change of variables, the DFT Y[k] can be
the symmetry property expressed as..
y[n] = y[2n –1 – n]. N −1 N −1

• The 2N point DFT Y[k] of the length-2N sequence y[n] is Y [k ] = ∑ x[n] W2knN + ∑ x[n] W − kn
2N W2kN(2 N −1)
n =0 n =0
thus … N −1
2 N −1 = W2−Nk / 2 ∑ x[n] (W2knNW2kN/ 2 + W2−NknW2−Nk / 2 )
Y [k ] = ∑ y[n] W kn
2N , 0 ≤ k ≤ 2 N − 1. n =0
n =0 N −1
⎛ π k (2n + 1) ⎞
• We can rewrite the above eqn. As…. X DCT [k ] = ∑ 2 x[n] cos ⎜ ⎟, 0 ≤ k ≤ 2 N − 1.
N −1 2 N −1
n =0 ⎝ 2N ⎠
Y [k ] = ∑ y[n] W2knN + ∑ y[n] W kn
2N
n =0 n= N • The Type 2N-point discrete cosine transform (DCT), X DCT [k ]
N −1
= W2−Nk / 2 ∑ x[n] (W2knNW2kN/2 + W2−NknW2−Nk /2 ) of the length-N sequence x[n] is obtained from the above
k /2
n=0 expression by extractingWthe : N samples of Y[k] and
2 N first

multiplying them with

N −1
⎛ π k (2n + 1) ⎞
X DCT [k ] = ∑ 2 x[n] cos ⎜ ⎟, 0 ≤ k ≤ N − 1. • Above eqns. Constitute a discrete cosine transform pair for
n=0 ⎝ 2N ⎠ the sequence x[n], where the first one is the Analysis eqn.
And the second one is the Synthesis eqn.
• Note that the samples of X DCT [k ] are real for a real
sequence x[n]. • A DCT pair may often be denoted as …
• The inverse discrete cosine transform (IDCT) of an N- DCT
point X DCT [k ] is given by … x[n] ↔ X DCT [k ].

1 N −1 ⎛ π k (2n + 1) ⎞
x[n] = ∑ α [k ] X DCT [k ] cos ⎜⎝ 2 N ⎟⎠ ,
N n =0
0 ≤ n ≤ N − 1.

• Where…
⎧1
⎪ , k = 0,
α [k ] = ⎨ 2
⎪⎩1, 1 ≤ k ≤ N − 1.

THANK YOU