Database Reference
In-Depth Information
Traditionally, energy histograms of DC coefficients are chosen to represent the
content of each frame. Since adjacent frames tend to be similar in content, only
Intra (I)- and Predictive (P)-frames are examined. The first step is to obtain the
DC coefficients in the compressed frames. As DC coefficients in the P-frames are
not readily available, frame conversion [
187
] or approximation methods [
188
] need
to be performed in order to extract the DC coefficients. In this work, the frame
conversion method [
187
] is selected as it introduces less error terms.
An energy histogram of a DCT coefficient is obtained by counting the number
of times an energy level appears in the DCT coefficient blocks of a DCT encoded
frame. For each of the luminance and chrominance components, one histogram is
created independently. The City block distance function is used to calculate frame
distance for component
i
, and it is defined as:
M
t
=
1
|
h
n
−
1
(
t
)
−
h
n
(
t
)
|
1
T
i
(
)=
D
i
n
(7.1)
where
h
n
−
1
(
1 and
n
respectively.
M
is the number of histogram bins. The histograms are usually zero padded and
aligned such that bins in the same position represent the same energy range.
T
i
is the
number of DC coefficients of component
i
in the frame, and it is used to normalize
the component frame distance
D
i
. Thus, the value
D
i
has an upper bound of 2,
which occurs when the energy ranges of
h
n
−
1
(
t
)
and
h
n
(
t
)
are the energy histograms of frames
n
−
do not overlap at all prior
to zero padding. The component frame distances are then combined to form the
overall frame distance
D
t
)
and
h
n
(
t
)
(
n
)
by averaging the three components:
1
3
∑
D
(
n
)=
D
i
(
n
)
(7.2)
i
∈{
Y
,
Cb
,
Cr
}
The presence of a sharp transition is indicated by a sharp impulse in the frame
distance function, and it can be detected with ease by thresholding. Alternatively,
D
(
)
is differentiated before thresholding is applied, and this reduces the difficulty
in threshold selection. In the case of a gradual transition, its presence is indicated
by a small peak with a width larger than 1. A sliding window is usually used to
accumulate frame distances caused by gradual transitions. However, since these
transitions can vary greatly in length, it is difficult to set the size of the window.
Even with an optimal window size, it is still possible miss a gradual transition due
to a small value of the accumulated frame distance.
n
7.2.2
Twin Window Amplification Method
The twin window amplification method (TWAM) was initially designed to enhance
gradual transitions. Before explaining this technique, we first need to discuss the
noise signals present in the frame distance function. During the MPEG compression
Search WWH ::
Custom Search