6th European Conference - Academic Conferences

Kesav Kancherla and Srinivas Mukkamala
ij ij ij
h = ( h , K , h )
(1)
L R
The next 11 functions are dual histograms represented with 8 × 8 matrices g d
i, j,where i, j = 1, . . . , 8,
d =−5, . . . , 5
nB
d
g = ∑δ ( d, dij(
k ))
(2)
ij
k = 1
Where δ(x, y) = 1 if x = y and 0 otherwise. For reducing the features only (i, j) ∈ {(2, 1), (3, 1), (4, 1),
(1, 2), (2, 2), (3, 2), (1, 3), (2, 3), (1, 4)} are taken
The next 6 functions capture inter-block dependency among DCT coefficients. The first function is
variation V
8 | Ir| −1 8 | Ic|
−1
∑∑ ∑∑
| d ( I ( k)) − d ( I ( k+ 1)) | + | d ( I ( k)) − d ( I ( k+
1)) |
ij r ij r ij c ij c
ij , = 1 k= 1 ij , = 1 k=
1
| I | + | I |
r c
Where Ir and Ic denote the vectors of block indices 1. . . nb while scanning the image by rows and by
columns, respectively
The next two functions capture the blockings of the frames
B
⎣⎢( M−1)/8 ⎦⎥ N ⎢⎣( N−1)/8⎥⎦
M
α α
∑ ∑| C8, i j − C8i+ 1, j | + ∑ ∑|
Ci,8 j −Ci,8j+
1|
α =
i= 1 j= 1 j= 1 i=
1
N ⎢⎣( M − 1)/8 ⎥⎦+ M ⎢⎣( N −1)/8⎥⎦
Where M and N are image height and width in pixels and ci, j are grayscale values of the
decompressed JPEG image, α = 1, 2
The last sets of features are co-occurrence matrix of DCT coefficients in neighboring blocks. The cooccurrence
matrix is calculated for values -2 to +2.
4.2 Markov features
From each image F (u, v), we obtain the following difference matrix along the horizontal, vertical,
diagonal and minor diagonal directions.
Fh( u, v) = F( u, v) − F( u+ 1, v)
Fv(,) u v = F(,) u v − F(, u v+
1)
Fd(,) u v = F(,) u v − F( u+ 1, v+
1)
Fm(,) u v = F( u+ 1,) v − F(, u v+
1)
Where F (u, v) is the image u, v gives the pixel location
In order to reduce the dimensionality we consider only the values [-4, +4] in these matrixes. Thus all
the values that is larger than +4 are set to +4 and the values that are smaller than -4 are set to -4.
From these we calculate the transition matrix as follows
146
(3)
(4)