DWT-Based Multisource Image Fusion Using
Spatial Frequency and Simplified Pulse Coupled
Neural Network
Nianyi Wang
1, 2
1. School of Information Science and Engineering, Lanzhou University, Lanzhou 730000, China
2. School of Mathematics and Computer Science Institute, Northwest University for Nationalities, Lanzhou 730000,
China
Email: xbmusailor@126.com
Yide Ma*
School of Information Science and Engineering, Lanzhou University, Lanzhou 730000, China
*Corresponding Author, Email: maydmayd@126.com
Weilan Wang
School of Mathematics and Computer Science Institute, Northwest University for Nationalities, Lanzhou 730000, China
Email: weilanchina@gmail.com
Abstract—In this paper, we present a new discrete wavelet
transform (DWT)-based multisource image fusion
algorithm using spatial frequency and a simplified pulse
coupled neural network model named spiking cortical
model (SCM). The multiscale decomposition and
multi-resolution representation characteristics of DWT are
associated with global coupling and pulse synchronization
features of SCM. Firstly, source images are decomposed
into low frequency sub-bands and high frequency sub-bands
by DWT. Secondly, considering human visual system
characteristics, two different fusion rules are used to fuse
low and high frequency sub-bands respectively. Maximum
selection rule (MSR) is used to fuse low frequency
coefficients. As to high frequency subband coefficients,
spatial frequency (SF) is calculated and then imputed into
SCM to motivate neural network rather than inputting
coefficients value directly, and then the time matrix of SCM
is set as criteria to select coefficients of high frequency
subband. The effectiveness of the proposed algorithm is
achieved by the comparison with existing fusion methods.
Index Terms—Multisource Image Fusion; Spiking Cortical
Model (SCM); Discrete Wavelet Transform (DWT);
Simplified PCNN; Spatial Frequency
I. INTRODUCTION
Image fusion extracts information from two or more
source images into a single composite image. The fused
image provides a more informative and comprehensive
description, and is more suitable for human visual
perception. There are many kinds of image fusion
methods. Along them, those methods that based on
multiscale decomposition (MSD) of source images
become more popular and important tools in recent years.
MSD methods decompose source images into high
frequency and low frequency subbands. Detailed and
coarse features remain in two types of subbands,
respectively [1]. New MSD methods are introduced in
image fusion, such as Curvelet [2], Ridgelet [3],
Contourlet, and Ripplet [4], etc.
In MSD domain, the discrete wavelet transform (DWT)
becomes one of the most popular methods. The DWT is
suitable for image fusion for the following reasons: (1) It
is a multiscale method that well suited to manage
different image resolutions. (2) The DWT allows image
decomposition in different levels of coefficients to
preserve more image information. (3) Coefficients
coming from different images can be combined to obtain
new coefficients, so that the information in original
images is collected. (4) Once coefficients are merged, the
final fused image is achieved through the inverse discrete
wavelets transform (IDWT), where the information in the
merged coefficients is also preserved [5]. In the DWT
based image fusion, the key step is to define a fusion rule
to create a new composite Multi-resolution representation.
Up to now, the widely used fusion rule is the maximum
selection (MS) scheme [6]. This simple scheme just
selects the largest absolute wavelet coefficient at each
location from input images as the coefficient at the
location in the fused image. However, as we know that
noise and artifacts usually have higher salient features in
the image, therefore, this method is sensitive to noise and
artifacts. Researchers widely discussed and proposed
some solutions to this problem [7].
As one of the third generation artificial neural
networks, Pulse coupled neural network (PCNN) is
another important image processing tool. It is a visual
cortex-inspired network characterized by global coupling
and pulse synchronization of neurons [8], and has been
widely applied in intelligent computing including image
JOURNAL OF MULTIMEDIA, VOL. 9, NO. 1, JANUARY 2014
doi:10.4304/jmm.9.1.159-165