nbul 2004
Kluwer
ping the
er Valley,
amization
p 64-76.
uation Of
Spectral
ience and
024.
‘Sub pixel
es”, Ph.D;
ty.
ping from
jan Area”,
,andgrebe,
erspectral
Electrical
tralia.
"Spectral
mpression
ernational
5,vol.4885
echt K.B.,
293, “The
interactive
ter data”,
sthods for
Symp. on
roc. SPIE
Analysis”,
ital Image
"oefficient
ssification
e Sensing,
'shold for
Journal of
V. 1992,
bers using
nmaries of
shop, JPL
3D WAVELET COMPRESSION TO MULTIPLE BAND REMOTE SENSING IMAGES
BASED ON EDGE RESERVATION
Qingquan LI" ** Qingwu HU n
‘Spatial Information and Network Communication Research and Development Center, Wuhan University, Wuhan, PR
China, 430079
School of Remote Sensing and Information Engineering, Wuhan University, Wuhan, P.R. China, 430079
KEY WORDS: 3D Wavelet Transformation, Remove Correlation, Edge Reservation, Image Restoration
ABSTRACT:
In this paper, a practical quasi-lossless compression concept and corresponding compression ratio and quality requirements are
proposed for the multiple band images database construction and web release application. The physical therapy of 3D wavelet
analysis to the multiple band images data compression is discussed and the fast 3D wavelet transformation model and algorithm to
the multiple band images is designed. The proposed algorithm is fully exploit the spectral and spatial correlation in the data. To
adopt to the local edge characteristics of multispectrum image, the 3D wavelet multiple band images compression technique route is
proposed based on the contour and edge feature of the multiple band images in the same region. And the remove correlation method
of the multiple band images contour feature keeping using 3D wavelet analysis and relative quantification coding methods to
different wavelet coefficient based on edge preservation is presented to code the multiple band images. The compression and
reconstruction experiment results to the 16-band images of the imaging spectrum sensor and the 36-band MODIS images can obtain
compression ratio over than 16 with PSNR to 42 and reach the quasi-lossless requirements, which show that this compression
technique can improve the quality of reconstruction images to the requirement of quasi-lossless with the high compression ratio.
1. INTRODUCTION
The authors propose that exparty pursuing compression ratio
and the quality of reconstruction images are not advisable. The
quasi-lossless compression technique is the best way for the
multiple spectrum images compression. What is called “quasi-
lossless " is that the gray standard deviation of the homologous
pixels between the original image and the restoration image
after reconstruction is less than the quantified noise(Zhou,1999
and HU, 2001). At the same time, the accuracy of pixels must
be less than the sensing imaging system's distortion. Thus, we
can satisfy the high ratio compression and ensure not to lóss the
image information.
In this paper, a practical quasi-lossless compression concept and
corresponding compression ratio and quality requirements are
proposed for the multiple band images database construction
and web release application. The physical therapy of 3D
wavelet analysis to the multiple band images data compression
is discussed and the fast 3D wavelet transformation model and
algorithm to the multiple band images is designed. The
proposed algorithm is fully exploit the spectral and spatial
correlation in the data. To adopt to the local edge characteristics
of multispectral image, the 3D wavelet multiple band images
compression technique route is proposed based on the contour
and edge feature of the multiple band images in the same region.
And the remove correlation method of the multiple band images
contour feature keeping using 3D wavelet analysis and relative
quantification coding methods to different wavelet coefficient
based on edge preservation is presented to code the multiple
band images. The compression and reconstruction experiment
results to the 16-band images of the imaging spectrum sensor
and the 36-band MODIS images can obtain compression ratio
over than 16 with PSNR to 42 and reach the quasi-lossless
requirements, which show that this compression technique can
improve the quality of reconstruction images to the requirement
of quasi-lossless with the high compression ratio.
In RS, GIS and DPS (digital photogrammetry system), one of
key technique is how to deal with the real time transmitting of
huge remote sensing data and how to build image database. And
Building digital libraries has become white hot in this era of
internet and the World Wide Web. In image databases, many
images must be stored and retrieved, and in data
communication applications, the image must be small enough to
be transferred quickly. The loss less coding based on statistics
has low compression ratio. Although the wavelet compression
and fractal compression will reach a high compression ratio,
they belong to degraded compression and need much more CPU
time. which affect these methods actual application in the field
of the remote sensing. The remote sensing images have higher
spatial resolution in wider coverage areas, and a number of
spectral bands, their accessibility is hindered by the size of
images. To alleviate these limitations, the image data should be
compressed.
The research of multiple spectrum images compression without
lossless can just reach the ratio about 3:1 and it can be not used
for the real applications(Zhang,1998). As one kind of sequence
images, the multispectrum images have strong correlation
among different frame or band. The researches of
multispectrum image compression focus on lossless
compression. The lossless compression of 224 bands AVIRIS
images (Huffman, 1994) obtains the compression ratio of 1.33-
1.50:1 and the 7 bands Landsat TM images reach 1.7-2.4:].
Memen,1994 propose the prediction tree for multispectrum
image compression with the resumption of neighbour band
image having the same prediction tree and then the prediction
tree can be used to remove redundancy among different bands.
All these compression method to multispectrum images are
based on stastic and lossless and they can not reach high
compression ratio to meet the real application.
* Corresponding author: Tel.:0086-27-87686512; Fax: 0086-27-87882661; Email: qqli@whu.edu.cn
55
