1. DATA FUSION
Data fusion is a good way that could contribute to implement a
good monitoring system.
Data fusion is analogous to the ongoing cognitive process used
to integrate data continually from their senses to make
inferences about the external world.
The multi-sensor data fusion is a method to combine data from
multiple (and possibly diverse) sensors in order to make
inferences about a physical event, activity, or situation
including those applications like automatic identification of
targets or analysis of battle field situations [1].
However, following facts have been described [2].
1. Combining data from multiple inaccurate sensors, which
have an individual probability of correct inference of less than
0.5, does not provide a significant overall advantage
2. Combining data from multiple highly accurate sensors, which
have an individual probability of correct inference of greater
than 0.95, does not provide a significant increase in inference
accuracy
3. When the number of sensors becomes large, adding
additional identical sensors does not provide a significant
improvement in inference accuracy
4. The greatest marginal improvement in sensor fusion occurs
for a moderate number of sensors, each having a reasonable
probability of correct identification
-Different levels of data fusion
1. Pixel-level fusion: At the lowest level, uses the registered
pixel data from all image sets to perform detection and
discrimination functions.
2. Feature-Level Fusion: combines the features of objects that
are detected and segmented in the individual sensor domains
3. Decision-Level Fusion: Fusion at the decision level (also
called post-decision or post-detection fusion) combines the
decisions of independent sensor detection/classification paths
by Boolean (AND, OR) operators or by a heuristic score (e.g.,
M-of-N, maximum vote, or weighted sum).
2. IMAGE FUSION
Main purpose of the image fusion is to increase both the
spectral and spatial resolution of images by combining multiple
images.
For doing image fusion in the best way we should pay attention
to 3 concepts [3].
2.1. Image registration
Image registration is the process that transforms several images
into the same coordinate system. For Example, for given an
image, several copies of the image are out-of-shape by rotation,
shearing, twisting so this process will focus on solving these
problems
2.2.Image resampling
Image resampling is the procedure that creates a new version of
the original image with a different width and height in pixels.
Increasing the size is called up sampling, for example on the
contrast, decreasing the size is called downsamplig.
2.3.Histogram matching
Consider two images X and Y. If Y is histogram-matched to X,
the pixel values of Y is changed, by a nonlinear transform such
that the histogram of the new Y is the as that of X.
3. IMAGE FUSION METH ODS
3.1. Pan-sharpening
The goal of pan-sharpening is to fuse a low spatial resolution
multispectral image with a higher resolution panchromatic
image to obtain an image with high spectral and spatial
resolution. The Intensity-Hue-Saturation (IHS) method is a
popular pan-sharpening method used for its efficiency and high
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B3, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
spatial resolution. However, the final image produced
experiences spectral distortion. HIS stands for Hue Saturation
Intensity (Hue Saturation Value), this method contain 3 steps:
First: The low resolution RGB image is up sampled and
converted to HSI space, Second: The panchromatic band is then
matched and substituted for the Intensity band, Third: The HIS
image is converted back to RGB space
3.2. PC Spectral Sharpening
We can use PC Spectral Sharpening to sharpen spectral image
data with high spatial resolution data. A principal component
transformation is performed on the multi-spectral data. The PC
band 1 is replaced with the high resolution band, which is
scaled to match the PC band 1 so no distortion of the spectral
information occurs. Then, an inverse transform is performed;
the multi-spectral data is automatically resampled to the high
resolution pixel size using a nearest neighbor, bilinear or cubic
convolution technique.
3.3. Wavelet theory for spatial fusion
In many cases we would wish to examine both time and
frequency information simultaneously¢ this leads to wavelet
transformation. wavelet transformation is a type of signal
presentation that can give the frequency content of the signal at
a particular instant of time what’s more it has advantages over
traditional Fourier methods in analyzing physical situation -
where the signal contains discontinuities and sharp spikes so
Wavelet algorithms process data at different scales or
resolutions. The continuous wavelet transform (CWT) is defined
as the sum over all time of the signal multiplied by scaled,
shifted versions of the wavelet function (¥),this part is shown
on Equation. 1
e
C(Scale, position) — | f (t)w(scale, position, t)dt (1)
—o
-DWT (Discrete wavelet transform)
It decomposes an image in to low frequency band and high
frequency band in different levels, and it can also be
reconstructed at these levels, when images are merged in this
method different frequencies are proceeded differently, it
improves the quality of new images so it is a good method for
fusion at the pixel level.
4. RESULT AND DISCUS SION
Here we had a focus on multi time's image fusion; these images
may be captured at different times. The object of the image
fusion here is to retain the most desirable characteristics of each
image to monitor moving object, We discussed different
algorithms for data fusion at this paper but we had focus on
Wavelet analysis for fusion of temporal images for monitoring
moving object.The principle of image fusion using wavelets is
to merge the wavelet decompositions of the multi times images
using fusion methods applied to approximations coefficients
and details coefficients.
-In first step ,we tried to select suitable wavelet form for fusion
so in our experiment, seven types of wavelet families are
examined: Haar Wavelet (HW), Daubechies(db), Symlets,
Coiflets ,Biorthogonal , Reverse Biorthogonal , Discrete
meyer(dmey) we tried to select the best form of wavelet based
on correlation with original image so Daubechies(dbl) was
selected because of good result.
Then we tried to select level of decomposition based on
wavelet theory, the maximum level to apply the wavelet
transform depends on how many data points contain in a data
set, so we examined selected level based on fusion result so we
used decomposition in two levels, it could give us the high
quality for fusion. The result was shown on Figure 4 that yellow
circles on the picture, shows path of moving object, so fusion
| €) (C) = =: MM o "CS gm]
Fo em fa) NY heel AR ee DO
^
