(IJCSIS) International Journal of Computer Science and Information Security,Vol.8, No. 8, November 2010
orthonormal bases
Ψ
mn
(x) obtained through translation anddilation of a kernel function
Ψ
(
x
) known as mother waveletas given in eqn. [2],… (2)Where, m,n are integers. To construct the mother wavelet
Ψ
(
x
), it is required to determine a scaling function
φ
(
x
)given in eqn. [3],… (3)Then, the mother wavelet
Ψ
(
x
) is related to the scalingfunction as in eqn. [4],… (4)where,The coefficients
h
(
k
) have to meet several conditions for theset of basis wavelet functions to be unique, be orthonormaland also have a certain degree of regularity.
C. Wavelet Transform for Signals
In two dimensional cases, the one dimensional wavelettransforms are applied along both the horizontal andvertical directions
()
x
φ
is a one dimensional real, sequenceintegral scaling function defined as in [5]
2,
()2(2)
j j jk
xxk
φ φ
= −
… (5)Translation
k
determines the position of this onedimensional function along the x- axis, scale
j
determine itswidth along x axis and
2
2
j
controls its height andamplitude. This one dimensional scaling function satisfiesthese conditions:
,
jk
φ
is orthogonal to its integer translates.
The set of functions that can be represented as aseries expansion of
,
k
φ
at low scale is containedwithin those at higher scale.So, the difference between any two sets of
,
jk
φ
isrepresented by a companion wavelet function
,
k
ψ
definedin eqn. [6],
2,
()2(2)
j j jk
xxk
ψ ψ
= −
… (6)
Then, the 2 dimensional DWT functions are thelinear products of scaling and wavelet functions
()
x
φ
and
()
ψ
yielding the eqn. [7] through eqn. [9].
(,)().()
H
xyxy
ψ ψ φ
=
… (7)
(,)().()
V
xyxy
ψ φ ψ
=
… (8)
(,)().()
D
yxy
ψ ψ ψ
=
… (9)where,
(,)
H
xy
ψ
,
(,)
V
xy
ψ
and
(,)
D
xy
ψ
are called thehorizontal, vertical and diagonal wavelets. Thus, DWT iswell localized and allows decomposition in three directions,namely, horizontal, vertical and diagonal respectively.
D. Features of Wavelets
In this application, the features are extracted using awavelet which belongs to a family of orthogonal wavelets.The mother wavelet (DB4), its corresponding scaling andwavelet functions and the decomposition filters are shownin Figure 3 and Figure 4 respectively.Fig. 3 wavelet extractionFig. 4 Decomposition of low-pass filter
h
φ
(-
n
) and high- pass filter
h
ψ
(-
m
)The DB4 scaling function is given by
02121222323
iiiii
ahshshshs
+ + +
= + + +
…(10)
[ ] [ ] [ ] [ ] [ ]
0123
2212223
aihsihsihsihsi
= + + + + + +
… (11)The Daubechies DB4 wavelet function is given by
02121222323
iiiii
cgsgsgsgs
+ + +
= + + +
… (12)
[ ] [ ] [ ] [ ] [ ]
0123
2212223
cigsigsigsigsi
= + + + + + +
… (13)
IV.
N
EURAL NETWORKS FOR SURFACE ROUGHNESSASSESSMENT
The roughness features extracted from the machinedimages, are fed as input to an ANN to predict the roughnessvalue R
t.
ANN consists of a number of elementary unitscalled neurons. A neuron is a simple processor, which cantake multiple inputs and produce an output. Each input intothe neuron has an associated weight that determines the‘‘intensity’’ of the input. The processes that a neuron performs are: multiplication of each of the inputs by itsrespective weight, adding up the resulting numbers for allthe inputs and determination of the output according to theresult of this summation and an activation function. Data isfed into the network through an input layer, it is processedthrough one or more intermediate hidden layers and finally
245http://sites.google.com/site/ijcsis/ISSN 1947-5500