You are using an outdated browser that does not fully support the intranda viewer.
As a result, some pages may not be displayed correctly.

We recommend you use one of the following browsers:

Full text

Mapping without the sun
Zhang, Jixian

l Multifrquency Po
ll- System. In: 2002
ng Symposium and
using, Toronto.
:velopment Kit. In:
Feng Xie, Yingying Chen, Yi Lin
The Research Center of Remote Sensing and Geomatic, Tongji University, Siping Road, Shanghai, China,
, Clipp, B., Engels,
iha, S., Talton, B.,
es, H„ Welch, G„
. Real-time video-
In: Proceedings of
PARCH 2007: 3D
Complex Architec-
KEY WORDS: SAR, Image Fusion, Pareto Optimum, Adaptive Multi-objective Optimization, Information Extraction, Image
Quality Evaluation
le Version 1.0. In:
:s and Technology
\mming Techniques
jl-Purpose Compu-
darris, M., Kriiger,
ey of General-Pur-
: Computer Graph-
Various image fusion algorithms have been developed to enhance the information of the image. However, few comprehensive
studies have been conducted for the information extraction of SAR with optical image. In this paper we focus on some typical fusion
algorithms based on multi-objective optimization was presented, which could achieve the optimal fusion indices through optimizing
the fusion parameters especially for extracting information and employ experimental testing to compare the performance. To judge
the performance of algorithms, we investigate both subjective and objective evaluation measures. Human evaluations of the fusion
results are also presented. Furthermore, we studied various image quality measures to evaluate optimizing algorithms objectively
under different decomposition levels, which include some standard quality metrics and other newly developed methods.
Performance evaluation experiments show that observers generally prefer the SiDWT pyramid with parameters using adaptive multi
objective optimization scheme for the realization of the Pareto optimum. The comparative tests prove the conclusion above; obtain
some valuable evaluation metrics and indicate using more decomposition levels will not necessarily produce better results under the
iter generated pic-
), pp. 311-317.
ssen, U., 2007. Fu-
>ptical imagery in
ilization and inter-
WG2 & WG7 Con-
and Optical Data,
ibul, Turkey.
, Stilla, U., 2006.
and challenges for
ngs - Radar, Sonar
of Radar Scatter-
rwood, Massachu-
ssion. In: Proceed-
hetic Aperture Ra-
>n curved surfaces.
Proceedings of the
cs and interactive
new Semi-empiri-
is heterogenity. In:
/ Geoscience and
ian Symposium on
SAR Image fusion is a valuable process in combining images
with various spatial, spectral and temporal resolutions to form
new images with more information which can be obtained
easily than that can be derived from each of the source images
[1]. In order to enhance the image information from SAR image
and to improve the abilities of the information analysis,
classification and the feature extraction, we focus on fusion
algorithms especially for the information extraction application
and employ experimental testing to compare their performance.
To judge the performance of algorithms, we investigate both
subjective and objective evaluation measures.
Various methods of fusion have the same objective, that is, to
enhance the interpretation of the SAR image. Different methods
have the given parameters, and different parameters could result
in different effects. In general, we establish the parameters
based on experience or the parameters adaptively change with
the image contents, so it is fairly difficult to gain the optimal
effects. If one image is as one information dimension or a
feature subspace, the procedure can be regarded as an
optimization problem in several information dimensions or the
feature space. A better result can be acquired through searching
the optimal parameters and discarding some in the processing.
So, a proper search strategy is very important for the
optimization problem. The better method was traditional
explored, and the most popular metrics is the mean square error
(MSE), which can not use for the case get an ideal or reference
image (that describes what the perfect scheme would produce)
is a very difficult task.
There are kinds of evaluation indices, and different indices
may be compatible or incompatible with one another, so a good
evaluation system of the procedure must balance the advantages
and disadvantages of different indices. The traditional solution
is to change the multi-objective problem into a single objective
problem using weighted method. The relation of the indices is
general nonlinear, and the method needs to know the weights of
different indices in advance. So it is highly necessary to
introduce multi-objective optimization methods based on Pareto
theory to search the optimal parameters in order to realize the
optimal results, by which the methods are more adaptive and
competitive because they are not limited by the given weights.
There are some multi-objective optimization algorithms include
Pareto Archive Evolutionary Strategy (PASE)[2], No
dominated Sorting Genetic Algorithm II (NSGA-II)[3],
Multiple Objective Particle Swarm Optimization (MOPSO)[4],
etc. A Lot of experiments with the two objective optimization
problems show that MOPSO has a better optimization capacity
and a higher convergence speed [4]. So we present an adaptive
multi-objective optimization to optimize the parameters for a
higher speed and better exploratory capabilities [5]. The
approach with adaptive multi-objective optimization is more
The paper focuses on fusion algorithms especially for SAR
with optical images applications and compares their
performance, which is organized as follows. Section 2 presents
a generic framework and describes the algorithms selected for
the study [6]. Some frequently used MSD (Multiscale-
Decomposition-Based) methods such as LPT, DWT, and
SiDWT are briefly reviewed [7]. Different alternatives and
optimization in the fusion procedure are then described,
especially expatiates adaptive multi-objective optimal
coefficient combining method in details. Then some standard
quality metrics and other newly objective quality measures are
introduced in Section 3. Further, Section 4 presents some
experimental results, subjective and objective quality measures
compare the performance of the selected algorithms. Finally,
conclusions are drawn in the Section 5.