IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India,2002
GENERATION OF OPTIMIAL QUANTIZATION TABLE FOR A RATE CONSTRAINED
VERSION OF JPEG USING GENETIC ALGORITHM
B.K.Mohan“, A.D.Devasthale^, and E.P.Rao*
* Centre of Studies in Resources Engineering, IIT, Bombay, Powai, Mumbai-400076, INDIA
? International Max Planck Research School, Max Planck Inst. of Meteorology, B-20146, Hamburg, Germany
* Department of Civil Engineering, IIT Bombay, Powai, Mumbai-400076, INDIA
bkmohan @csre.iitb.ac.in
KEY WORDS: Remote Sensing, Image Compression, JPEG, Genetic Algorithms
ABSTRACT:
Image compression using the JPEG technique has been popular for a long time with imaging device manufacturers, website
developers, educational content developers and more recently the remote sensing community. The technique has been well
understood, and several hardware encoders and decoders are available to save in /retrieve from JPEG format. With the increase in
spatial and radiometric resolution of spaceborne imaging sensors, image compression is a serious consideration and countries like
India and France are embedding image compression technology in hardware for real time onboard application. The JPEG scheme
has one limitation in its original definition in that it is not possible to specify a chosen compressed image size by the user. This
paper describes an attempt at the use of genetic algorithms for generation of JPEG quantization tables using which the desired
compression ratio can be achieved.
1. INTRODUCTION
1.1 Image compression in remote sensing
Increase in the resolutions of remote sensing imageries has
resulted in the large volume of data to be handled, processed
and stored. The on-board or online compression has also
become very important. Since lossless compression offers very
small compression ratio, we have to opt for lossy methods.
Among various lossy compression methods available, the most
widely used is JPEG method. JPEG method is based on the
Discrete Cosine Transform followed by the quantization and.
entropy coding. The extent of compression achieved depends
upon the coarseness of quantization of the transformed
coefficients. Since JPEG offers compression based on quality
factor, which ranges from 0 to100, we cannot specify the exact
compression ratio or bit-rate while compressing the images. A
quality factor is the option available for this purpose that
indirectly leads to generating larger or smaller compressed
image size. This solution is not often satisfactory since this may .
result in over-compression causing excessive loss of
information or under-compression causing overloading of the
data downlink. Very approaches are published in literature on
the use of optimisation techniques that can produce close to
desired size compressed images. JPEG makes use of default
quantization table for corresponding quality factor. Several
approaches have been already used to design the quantization
tables, which will result in specified compression ratio.
An attempt is made to design an algorithm for optimal
quantization table to get specified compression ratio for remote
sensing images using genetic algorithm. Algorithm makes use
of error occurred in getting desired file size and image
distortion (expressed in terms of Mean Squared Error) as a
fitness criteria.
2. JPEG COMPRESSION
One of the earliest standards in data compression is the Joint
Photographic Experts Group (JPEG) defined standard, better
known as JPEG compression. This involves three main steps:
a) Discrete Cosine Transform (DCT) of the image
b) Quantization of the DCT coefficients
¢) Coding of the quantized coefficients
Usually the input image is divided into blocks of size 8x8 (Fig.
1), and each block is separately processed. The degree of
compression is controlled by a quality factor which determines
how many of the low magnitude transform coefficients may be
discarded, or how much compression can be achieved. Greater
the value of the quality factor, better the quality of
reconstruction at the cost of reduced compression factor.
here is no way in the standard formulation for the user to seek
a desired compression ratio at the best available quality. It is
known from the JPEG standard that image compression and
also associated loss occur during quantization of DCT
coefficients. The quantization step is executed by dividing
every element of the DCT block by a corresponding element of
a quantization block or table. A default quantization table is
specified within the JPEG standard, and a provision is made for
using custom tables.
The JPEG committee has released a new standard known as
JPEG-2000 based on wavelet transform, which is said to have
overcome all the limitations of JPEG. But since JPEG is
already being in use from so many years and its real time
implementation is available, the usage of JPEG will continue
till the real time implementation of JPEG-2000 standard
becomes available for remote sensing images.
