Èñòî÷íèê: «Èíôîðìàòèêà è êîìïüþòåðíûå òåõíîëîãèè – 2010» — 2010 / Ýëåêòðîííûé ñáîðíèê ìàòåðèàëîâ ìåæäóíàðîäíîé íàó÷íî-òåõíè÷åñêîé êîíôåðåíöèè ñòóäåíòîâ, àñïèðàíòîâ è ìîëîäûõ ó÷¸íûõ «Èíôîðìàòèêà è êîìïüþòåðíûå òåõíîëîãèè – 2010». — Äîíåöê, ÄîíÍÒÓ — 2010
UDC 004.932.2:004.627
Fractal
compression algorithms
in context of medical images` procession
Anastasova E.A., Belovodskiy V.N.
E-mail: anastasova.k@gmail.com
Abstract
The topicality of medical images’
procession was covered. There was considered one of the most essential problems
of telemedicine furthering, development and applying – storage and transmission of graphics.
Possible methods of images’ procession were discussed. The modifications of
compression algorithm, that provided high compression efficiency of information
and quite small
decompression error values, were analyzed.
Introduction
Images are widely used in different fields
both in casual life either in narrow terms of science. Medical images are almost all grayscale ones.
It is necessary to notice that
only the part of whole image is worth analyzing. There is
a need for huge volumes of memory to store and qualitative process the high
quality images. As a result some difficulties of real-time processing occur. According
to medical images some characteristics of the algorithm can be named: compress
ratio is high, compress time vary, but decompress one seeks to minimize, output
error is minimal.
Problem statement
The object of the paper is analysis of
exists treatments and methods those provides quite low error after decompress to
allow next assay, high compress ratio, short time for decompression.
Solutions to the problem
It is necessary to discuss the
milestones of basic fractal algorithm. It implies a partition of origin image on
domains and ranks. After this step domains are sorted out for each rank (the
domain is compressed in each orientation to the size of rank and the best
values of coefficients of process using the method of least squares.
Due to process and usage specificity
of medical images the rule “once to compress and ofttimes
decompress” can be named. In this
connection the method offered by Vatolin is fully applicable [1]. The author supposes
to allocate significant areas of an image, to use different compression ratio
depending on characteristics of image’s parts. The half automatic systems are used
to choose the areas. A domain is chose from region, that approximate the rank quite
well and isn’t much worse during the compression. The mean distance from the correction is used
as a measure of optimal block.
In medical
images the main interest is only its fragments, not a whole one. It’s offered to
compress only the fragments tacking into account their features to save a high
quality of important parts of image while ensuring the highest possible degree
of compression. For this an image is divided into disjoint parts tacking into account
their informational importance or morphological structure on purpose of further
compression of each of fragments using the preferable algorithm in terms of the ratio of two main
characteristics “compression rate” and “quality”. To solve this problem, introduce the concept of a mask that shows the
location of one or more areas of interest. This area can be allocated on the
basis of structural features - images heterogeneity in terms of there are so-called
constant regions present, where all pixels have identical or close to the value
of shade, and areas with lots of small parts, where the neighboring pixels
differ in color . In this case, use the statistical characteristics, defined by luminance
histogram. In particular, we can analyze images based on the values of the average
entropy:
here L - quantity of colors gradations,
- luminance histogram,
- random
value, that features luminance of picture elements.
The values of entropy for the whole
image and for each area separately are calculated, and these values are
compared. A sign of small details in the field is increasing of the entropy values
compared to the entropy of the entire image. This region (one or several) of the
first level is divided into areas of the second level [2].
At the stage of compression selections can be used different algorithms. We describe the best ones.
The most common
modification of the basic fractal algorithm is the FE-algorithm. The comparison of five
characteristics that describe the domain and rank blocks is helps to reduce the
computational cost of FE-algorithm. They are compared in the
beginning of searching. These characteristics are:
standard deviation, asymmetry, inter-pixel contrast, 
coefficient, which characterizes the
differences between pixel values and the value of central pixel, the maximum gradient - the maximum of horizontal and vertical gradients. The rank of unit’s characteristics
vector is calculates when the unit is processed, and then the distance between
the characteristic vector of the rank and the characteristic vector of each
domain is calculated. Procedure for selection of domains is a kind of filter, which
significantly limits the number of domains that are moving [3]. Pearson's correlation coefficient can also be
used to optimize search of best domain. The better the actual dependence of
R D is approximated by the linear; the closer to
,
i.e. the domain contrast should be
higher than rank contrast grade [4]. Algorithms of this kind can be
efficiently implemented using Unsupervised Kohonen maps [5].
It
is important to note, that separate place is
occupied by the algorithms using basically discrete pseudocosine transformation
(DPCT). Such algorithms are characterized by performances rather approximated to
that the JPEG method shows. Estimates of the computational costs show that the algorithm based on DPCT
not inferior to JPEG, including the computational complexity value [6].
Conclusion
Thus, there is an analysis of approaches and
algorithms for image processing. Most interesting are the allocation
of significant areas and the application of specific compression algorithm for
each one. This allows us to achieve significant compression ratios and retain all
the important parts in satisfactory condition.
The aim of future work is to create applications based on existing
approaches, their approbation on real images, and the modernization of
algorithms.
List of sourses:
1.
Vatolin, D.S. Increasing the degree of
compressing fractal image compression by specifying the quality of image areas
/ International Conference Graphicon 1999,
2.
Zulema, E.S. Adaptiva image compression method / News of Khmelnytsky national university ¹ 2 – 2010
3.
Bublichenko, A.V. Algorithms for image compression: a
comparative analysis and modification / A.V. Bublichenko, V.N. Belovodsky / Qualification Masters
work. - 2008.
4.
Ilyushin, S.V. Fractal compression of
telemedicine images / S.V. Ilyushin, SD Light / "Telecommunications», ¹ 4 - 2009.
5.
Prokhorov, V.G. Using Kohonen maps to accelerate
fractal image compression / V.G. Prokhorov / Applied Software, ¹ 2. - 2009. - S. 7.
6.
Umnyashkin, S. V. Mathematical methods and algorithms
for digital image compression using orthogonal transformation / S. V. Umnyashkin / Abstract. - 2001.