Digital image compression technology Research and Development
Tian Yong 1, DING Xue-Jun 2
(1. Northeastern University, embedded systems Neusoft Institute of
Information Engineering, Liaoning, Dalian 116023; 2. Northeast University of
Finance and Information Engineering, Liaoning, Dalian 116023)
Abstract: Digital image compression technology for digital image information
on the network for rapid transfer and real-time processing is of great
significance. This article describes several of the most important of the current
Compression algorithms: JPEG, JPEG2000, fractal image compression and
wavelet transform image compression, summed up the advantages and
disadvantages of their development prospects. A brief introduction and then
arbitrarily shaped visual object coding Algorithm for the status quo, and pointed out that this algorithm is a high
compression ratio resulting image compression algorithms.
Keywords: JPEG; JPEG2000; fractal image compression; wavelet transform;
arbitrarily shaped visual object coding CLC number: TP3 Abstract: A Article ID :1672-545X (2007) 04-0072-04 With
the multimedia technology and communication technology continues to evolve,
multimedia entertainment, the letter
Interest rates have kept the information highway data storage and
transmission should be set higher
Demand, but also to the limited bandwidth available to a severe test, especially
those with large data
The amount of digital image communication, more difficult to transport and
storage, which greatly restricted the image pass The development of the letter, so the image compression technology has been
more and more attention. Image compression The purpose is to shrink the original image of a larger less bytes exhausted
expression and transmission,
And asked to recover a better quality image. The use of image compression
that can reduce the map
Such as the burden of storage and transmission, so that the image on the
network to achieve fast transfer and real-time Office Li.
Image compression techniques can be traced back to 1948, raised the number
of television signals
Word-oriented, and today has 50 years of history . During this period there
have been many
Kinds of image compression coding method, especially in the late 20th century,
80 years later,
Because the wavelet transform theory, fractal theory, artificial neural network
theory, visual simulation of Li
On the establishment of image compression technology has been an
unprecedented development, in which fractal image Image compression and wavelet image compression is the current hot
research. In this paper, present, the most widely Pan-use image compression algorithms are reviewed and discussed their
advantages and disadvantages, as well as hair Show prospect.
1 JPEG compression
Responsible for developing the still image compression standard "Joint
Photographic Expert Group" (Joint
Photographic Expert Group, referred to as JPEG), was formed in January
Adaptive DCT-based JPEG technical specifications of the first draft, followed
Changes to the 1991 formation of the draft international standard ISO10918,
and after one year into the
As an international standard, called JPEG standard. 1.1 JPEG compression principle and characteristics of JPEG algorithm in the first block of the image processing, generally divided
into each other without re -
The size of stacked blocks, and then every one to two-dimensional discrete
cosine transform (DCT). Variable
After changing coefficients are not relevant, and the coefficient matrix of the
energy concentrated in the low-frequency area, the root Quantization table, according to quantify, quantify the results of retained
low-frequency part of the coefficient, remove the The high-frequency part of the coefficients. The quantized coefficients by
Zigzag scan re-organization, Ran
After Huffman coding. JPEG features are as follows: Advantages: (1) the formation of international standards; (2) has a mid-range
and high bit-rate on the
Good image quality.
Disadvantages: (1) Because of the image block, at high compression ratio
when you have a serious
Blocking effects; (2) factor to quantify is the lossy compression; (3) the
compression ratio is not high, small
At 50 .
JPEG compressed image box effect occurs because: Under normal
circumstances the image
The signal is highly non-stationary, it is difficult to characterize the process of
using Gauss, and the image
Some of the mutant structure, such as the edge information than the image
smoothness important, with the cosine-based
Non-linear approximation for image signals the result is not optimal .
1.2 JPEG compression status and prospects of the study 
JPEG at high compression ratio for cases arising under the box effect,
decompresses the image than the Poor, in recent years made a lot of improvement methods, the most effective is
the following two ways:
(1) DCT zerotree coding
DCT block zero-tree coding the DCT coefficients in the composition of log2N
After the zero-tree coding scheme used to encode. Compression ratio in the
same circumstances, the
PSNR value higher than the EZW. However, in the case of high compression
ratio, the box effect is still
DCT the Achilles heel of zero-tree coding. (2) layer-type DCT zerotree coding The algorithm for the DCT transform the image will be low-frequency blocks
together, doing anti -
DCT; pairs of newly acquired images do the same transformation, and so it
goes, until satisfied
Request until the. Then Layered DCT transform and zero-tree arrangement of
the coefficients over zero
JPEG compression one of the biggest problem is that at high compression
ratio when you have a serious Blocking effects, so future research should be focused on solving the resulting
Blocking effects, taking into account the human visual system with the
combination of compression. 2 JEPG2000 compression
JPEG2000 is a ISO / IEC JTCISC29 standards group responsible for
The new still image compression standard. One of the biggest improvements
is that it uses wavelet transform on behalf of the
For the cosine transform. In March 2000 the Tokyo conference to identify a
color still image
Encoding a new generation of image compression standard-JPEG2000 coding
2.1 JPEG2000 compression principle and characteristics of
JPEG2000 Codec encoder and decoder block diagram in Figure 1 .
Encoding process is mainly divided into the following processes:
pre-processing, core processing and bit Stream organization. Pretreatment section includes image segmentation, DC
level (DC) displacement and sub -
Received date: 2007 - 01 - 16
Author: Takada (1975 -), male, Master, Lecturer, major research interests
include digital image processing, integrated circuit design, etc.; DING Xue-jun
(1978 -), female, master's degree, teaching assistants, the main research
direction for information
And network security, digital image processing. The amount of change. Core
processing in part by the discrete wavelet transform, quantization and entropy
coding form. Bit
Stream part of the organization include zoning, code blocks, layers and
JPEG2000 format image compression ratio can be based on the current JPEG
Further increased by 10% to 30%, and the compressed image is even more
delicate smooth. For
The current JPEG standard, in the same compressed stream can not provide
Lossless compression, while in JPEG2000 system, by selecting the
parameters, be able to image
For lossy and lossless compression. Now the network is based on JPEG
"Block" transmission, while the JPEG2000 image format support for
progressive transmission, which use the Households do not receive the entire image compression bit stream. Since
JPEG2000 uses wavelet technology Surgery may be random access to some image region of interest (ROI) of the
compressed stream, right
Compressed image data transmission, filtering and other operations .
Figure 1 JPEG2000 compression and decompression of the overall process
2.2 JPEG2000 compression prospects JPEG2000 standard applies to a variety of image compression coding. Its
Will include the Internet, fax, printing, remote sensing, mobile communications,
medical, digital libraries
And e-commerce . JPEG2000 image compression standard will become the
21st century, the main
Flow still image compression standard. 3 Wavelet Image Compression 3.1 Principle of Wavelet Image Compression Wavelet Transform for image coding basic idea is that the image according to
Tower fast wavelet transform algorithm for multi-resolution decomposition. The
specific process is:
First, the image multi-level wavelet decomposition, then the amount of the
wavelet coefficients of each layer , And re-pairs of the quantized coefficients encoded. Wavelet image
compression is the current image compression
One of the hot shrinkage has been formed based on wavelet transform of
international compression standards, such as the MPEG-4 standard, and as mentioned above the JPEG2000 standard .
3.2 Wavelet Image Compression Status and Prospects At present the three highest levels of wavelet image coding are embedded
wavelet zero -
Tree Image Coding (EZW), hierarchical tree distribution of the sample image
coding (SPIHT) and
Expand image compression coding (EBCOT). (1) EZW coder 
In 1993, Shapiro introduced a wavelet "zero-tree" concept, by defining
POS, NEG, IZ and ZTR symbols are four kinds of recursive spatial wavelet tree
coding, there are
Effective to remove the encoding of high-frequency coefficients, which greatly
improved the coding of wavelet coefficients Efficiency. This algorithm is used to quantify and embedded progressive
coding mode, the algorithm complexity Low. EZW algorithm for breaking the long-term deep faith in the field of
information processing criteria: highly efficient compression
Reduction encoder must pass the highly complex algorithms to get, so EZW
Data compression device in the history of a landmark decision.
(2) SPIHT encoder 
By Said and Pearlman proposed a collection of hierarchical wavelet tree
(SPIHT) is the use of space partitioning tree hierarchical approach to
effectively reduce the bit-plane
Coded symbol set size. EZW, compared with, SPIHT algorithm is constructed
by two kinds of non -
The same type of space zero-tree, to make better use of the wavelet
coefficients of the amplitude attenuation. Like with the EZW encoder, SPIHT encoder algorithm complexity low, the
Is also an embedded bit stream, but the encoder performance than EZW
(3) EBCOT encoder 
Optimized cut-off point of the embedded block coding (EBCOT) first wavelet
Each sub-band is divided into a relatively independent of the code block, and
then use a hierarchical optimization
Truncation algorithm to the code block is encoded to generate compressed
stream, the results of image compression Reduction stream is not only scalable but also has resolution of SNR can be
Extension, you can also support the random image is stored. In comparison,
EBCOT algorithm complexity compared with EZW and SPIHT are referred to
With its high compression performance than SPIHT increased slightly.
Wavelet image compression is considered the most promising One of the image compression algorithm. Wavelet image compression
research has focused
Wavelet coefficients in the coding issues. In later work, Should fully consider the human visual system to further improve the
To improve the image quality. And consider the wavelet transform and other
Method combination. For example, with the combination of fractal image
compression is the current
A research focus .
4 Fractal Image Compression
In 1988, Barnsley proved by experiments fractal image compression ratio can
Classical image coding technology, high compression ratio of several orders of
magnitude. In 1990, Barnsley
Students AEJacquin making a partial iterated function system theory, will bring
the use of fractal
In image compression on the computer automatically possible. 4.1 The principle of fractal image compression Fractal compression using mainly self-similar characteristics, through the
iterated function system (Iterated
Function System, IFS) to achieve. The theory is based on iterated function
Theorem and the collage theorem.
Fractal image compression the original image is divided into several
sub-images, and then each
Sub-image corresponds to an iterated function, sub-images to iterated function
storage, iterative letter
The number of the more simple, the greater the compression ratio. The same
decoding as long as the transfer out of each sub-map Like the corresponding iteration function iteration, it can be restored out of the
original sub-image, from the
Obtained the original image .
4.2 several major fractal image coding techniques  With the fractal image compression technology, an increasing number of
algorithms have been proposed,
Based on the different fractal characteristics can be divided into several major
(1) The size of encoding method
Size coding method is based on fractal geometry, the use of small-scale
Curve length method, similar to the traditional sub-sampling and interpolation
methods, the main non -
Scale with the encoding method is that the introduction of the idea of fractal,
scale with the image
The complexity of the various components of the different change.
(2) The method of iterated function system Iterated function system method is the most studied, the most widely used as a
Shaped compression technology, it is a patchwork of human-computer
interaction technology, it is based on the natural world map Like in the prevailing global and local autocorrelation features, search for such
Shot between the expression, that is, affine transformation, and by storing a
small amount compared with the original image data Affine coefficients to achieve the purpose of compression. If you find a simple
affine transformation Effective, then the iterated function system can achieve
very high compression ratio.
(3) A-E-Jacquin fractal program
A-E-Jacquin fractal program is an automated block-based fractal
Image compression scheme, it is also looking for a mapping process, but are
looking for right
As the domain is divided into blocks of the image after the partial and local
interests. In this scenario,
There are still some redundancy can be removed, and its decoding the image
there is a clear
4.3 The prospects for fractal image compression  Although the fractal image compression in the field of image compression is
not dominant, but the
Fractal Image Compression take into account both local and partial, but also
take into account the relevant part and the whole Nature, suitable for self-similar or self-affine image compression, but in nature
there is a large number of
Self-similarity or self-affine geometry, so its scope is broad. 5 other compression algorithms
In addition to these several common image compression methods, there are:
NNT (number theory
Transform) compression, the compression method based on neural networks,
Hibert scanned image compression side France, adaptive polyphase subband compression method, etc. This is not to
repeat. Following is a brief
In recent years, several arbitrarily shaped texture coding algorithm [10 ~ 13].
(1) The shape-adaptive DCT (SA-DCT) algorithm
SA-DCT to an arbitrarily shaped visual objects into the image blocks, each
To DCT transform, which implements a similar shape-adaptive Gilge DCT [10,
Transform an effective transformation, however, it Gilge DCT transform lower
SA-DCT also has shortcomings, it is the pixel onto one side of the border with
the rectangle relative to flat
Qi, so some spatial correlation may be missing, so further out DCT transform,
There is a greater distortion of the [11, 14, 15].
(2) Egger method
Egger and others [16, 17] proposed a wavelet applied to objects of arbitrary
For the program. In this scenario, the first visible line of pixels onto the object
and bounding box
The right boundary phase flush position, then the usefulness of each line pixel
Change, followed by a further direction of the wavelet transform. This program,
take advantage of a small
The local characteristics of wavelet transform. However, the program also has
its problems, such as might lead to Play an important part of the high-frequency part of the merger with the
boundary, can not guarantee that the distribution coefficient of each of the
The same inter-correct phase, as well as the second direction may cause the
Continuous and so on.
(3) The shape-adaptive discrete wavelet transform (SA-DWT)
Li et al proposed a novel encoding of arbitrary shape of the object, SA-DWT
Coding [18 ~ 22]. The technology includes the SA-DWT and zerotree entropy
(ZTE), and embedded wavelet coding (EZW). SA-DWT is characterized by:
SA-DWT coefficients after the count of visual objects of arbitrary shape with
the original number of pixels The same; wavelet transform of the spatial correlation, regional attributes, and
sub-band between the Zi Xiang Similarity, in the SA-DWT can be well manifested in; for the rectangular area,
SA-DWT and the same as the traditional wavelet transform. SA-DWT coding
technology to achieve
Have been new multimedia coding standard MPEG-4's for any shape of a
Rationale used by the encoding. In future work, you can fully utilize the human visual system to the image edge
Edge of some of the more sensitive features, try to image objects of interest to
Some of its edges, internal textures and objects other than some background
section according to different Compression ratio of compression, so that compressed images can achieve
greater compression ratio and more
Add to facilitate transmission. 6 Summary
Image compression technology research for decades and has made great
achievements, but there are Much to be desired, it is worth further examination. Wavelet image
compression and fractal image compression
Reduction is currently a hot research, but both also have their own
shortcomings, in the future work, Should be with the human visual properties. In short, image compression is a
Promising area of research, a breakthrough in this area of life for our
information and communication To the development of far-reaching implications.
Tian Qing. Image compression techniques [J]. Police Technology, 2002, (1): 30
ZHANG Hai-yan, Wang Mu, et al. Image compression technique [J]. Journal of
System Simulation, 2002, 14 (7): 831 - 835.
Zhang Zongping, Liu Guizhong. Wavelet-based video image compression
research [J]. Journal of Electronics,
2002, 30 (6): 883 - 889.
Zhou Ning, Tang Xiaojun, Xu Park. JPEG2000 image compression standard
and its key algorithm [J].
Modern electronic technology, 2002, (12): 1 - 5.
WU Yonghui, YU Jian-xin. JPEG2000 image compression algorithm overview
and network application prospects [J].
Computer Engineering, 2003, 29 (3): 7 - 10.
J MShaprio. Embedded image coding using zerotree of wavelet coefficients [ J]. IEEE Trans. On Signal Processing, 1993, 41 (12): 3445 - 3462. A Said, WA Pearlman. A new fast and efficient image codec based on set partitioning in hierarchical trees [J]. IEEE Trans. on Circuits and Systems for Video Tech. 1996, 6 (3): 243 - 250.
D Taubman. High performance scalable image compression with EBCOT [ J]. IEEE Transactions on Image Processing, 2000, 9 (7): 1158 -- 1170.
Xulin Jing, MENG Li-min, Jianjun. Wavelet image compression with branches
in comparison and applications.
China's cable TV, 2003, 03/04: 26 - 29.
M Gilge, T Engelhardt, R Mehlan. Coding of arbitrarily shaped image segments based on a generalized orthogonal transform [J]. Signal Processing: Image Commun., 1989, 1 (10): 153-180.
T Sikora, B Makai. Shape-adaptive DCT for generic coding of video [J]. IEEE Trans. Circuits Syst. Video Technol., 1995, 5 (1): 59-62. T Sikora, S Bauer, B Makai. Efficiency of shape-adaptive 2 - D transforms for coding of arbitrarily shaped image segments [J]. IEEE Trans. Circuits Syst. Video Technol., 1995, 5 (3): 254-258.
E Jensen, K Rijk, et al. Coding of arbitrarily shaped image segments [C]. Proc. Workshop Image Analysis and Synthesis in Image Coding, Berlin, Germany, 1994: E2.1-E2.4.
M Bi, SH Ong, YH Ang. Comment on "Shape-adaptive DCT for generic coding of video "[J]. IEEE Trans. Circuits Syst. Video Technol., 1996, 6 (6): 686-688.
P Kauff, K Schuur. Shape-adaptive DCT with block-based DC separation and Delta DC correction [J]. IEEE Trans. Circuits Syst. Video Technol., 1998, 8 (3): 237-242.
O Egger, P Fleury, T Ebrahimi. Shape-adaptive wavelet transform for zerotree coding [C]. Proc. Eur. Workshop Image Analysis and Coding for TV, HDTV and Multimedia Application, Rennes, France, 1996: 201 -- 208.
O Egger. Region representation using nonlinear techniques with applications to image and video coding [D]. Ph.D. dissertation, Swiss Federal Institute of Technology (EPFL), Lausanne, Switzerland, 1997. S Li, W Li, et al. Shape adaptive vector wavelet coding of arbitrarily shaped texture [S]. ISO / IEC JTC/SC29/WG11, MPEG-96 - m1027,