From all the senses granted unto us by Mother Nature it is the sense of vision, mediated by the eye, which supplies us with the most comprehensive and accurate picture of the surrounding world. In our high-tech day and age we can study different objects not only by looking at them directly, but also by perusing their photos, movies, video recordings and TV programs. And right from the start experts and engineers dealing with these various means of visual communications were confronted with the problem of improving the quality of the graphic images thus obtained. Until recently the problem was solved by developing better and better optical systems. But the advent of computers produced a revolution also in optics, caused by the introduction of digital methods. Some of the problems and achievements in this challenging field of modem research have been reviewed by a prominent expert in the field, Corresponding Member of the Russian Academy of Sciences, Viktor Soifer, Director of the Institute of Image-Processing Systems of the RAS Research Center in Samara (Central Russia).
First publications on the digital image processing appeared on the scene back in the 1960s in connection with research in astronomy, nuclear physics, biophysics and radio physics and were chiefly associated with ways of improving the input-output systems. As it later turned out, optical fields (images) could be represented as blocks of data in computer memory. This opened up new possibilities of their processing and provided the basis for the formation of new areas of research at the junction of informatics and optics, above all the field of computer applications wherein Russian scientists were the pioneers.
Here one should mention first of all the RAS Institute of Problems of Data Transmission-a pioneer in the field of image processing, the RAS Computer Center and the "Cybernetics" Scientific Council which provided an outstanding contribution to solving the problems of image recognition. The RAS Institute of Radio Engineering and Electronics, working in conjunction with the COMETA R&D Center and the TsKB- PROGRESS State Aerospace R&D Center were dealing with the problems of improving data collection and processing from space research while the RAS Institute of Mathematical Modelling was developing a theory of differential diagrams which was instrumental in dealing with problems of diffraction lying at the basis of computer optics. One should also mention here the contributions to these studies provided by the Lebedev Institute of General Physics, the RAS Institute of Optical-Neuron Technologies, the Institute of Automatics and Electrometry of the RAS Siberian Branch and the Physics Department of the Moscow State University which were all involved in computer optics research in those years. In 1993 they were joined by the RAS Institute of Image-Processing established in Samara.
These studies by Russian scientists have been largely linked with the obtaining, processing, storage and studies of images received from aerospace probes used for the remote monitoring of the Earth and which are now broadly used in meteorology, geology and mapping, oceanology, geobotanics, farming and forestry, in dealing with problems of land utilization and hydropower development, environmental monitoring, and so on and so forth. On the basis of such imagery, for
example, one can build what specialists call a computer and space-distributed data model of a territory under investigation. This, in its turn, helps keep check on the various local technogenic and natural processes, prognostication of economic development, social development and the state of the environment.
A model of this kind is being built up in stages, with a growing scale and complexity of the pursued objectives. This, as a rule, is done at three scales of magnitude: fine (review) - 1:100,000-1:200,000; medium (plan) - 1:10,000-1:25,000 and large (cadaster) - 1:500-1:2,000. The studies are conducted in two "time versions": statistical, when data changes accrue over months and years, and a "dynamical" one in which data are "updated" on a daily and even hourly basis. Every stage broadens the range of the pursued geo-analytical problems and requires a "higher quality level". That is why, in addition to aerospace imagery, experts also use for the building of such models other sources of information such as cartographic materials, data on coordinates measurements with the help of the satellite Global Positioning System (GPS), and so on.
The systems of remote probing of the Earth from space have a history of about 40 years. Today probes of this kind are quite common and differ by their complexity, the range of the tasks involved, degree of automation, operational and technical parameters. The "simplest" of them - with photographic image registration-feature high spatial resolution, but do not meet the modern requirements (1-5 m and more). Attempts to deal with the situation have caused a radical revision of the principles of building and functioning of such systems, their considerably high complexity involving the use of algorithms of in-depth mathematical image processing and analysis.
The traditional mechanical scanners have been replaced with optico-mechanical systems with linear or solid-state transistor sensors. These are more economical and reliable and, used in combination with high-quality optics, ensure the required resolution. And boosting the resolution serves to sharply increase the volume of data which makes it possible to pass from a "blanket" sounding of the Earth's surface to selective mapping of chosen areas. This, in its turn, dictates a very careful approach to the planning of the system of video measurements, expanding the base of cartographic data and for the onboard image processing for the purpose of coordinates "matching". After that the chosen areas are submitted to one more operation before they are transmitted to the Earth. This operation is known as "data compression" which helps to reduce the volume of communications by several times. Similar data "compressions" are also performed at ground receiving stations as a result of which we can use more effectively the capacity of data storage facilities for imagery of this kind.
All of the above advantages, however, also have a negative side. It turns out that with a high resolution and rate of video data transmission we get appreciable distortions of the optical signal. This snag cannot be overcome without developing some compensatory systems.
This thing is that the processing of images obtained with the help of aerospace remote sounding technology is subdivided into what we call "preliminary" and "final". The former includes all of the above techniques and ends us with the placing of imagery in the database for the subsequent "topical" uses thereof. The latter is used in dealing with some concrete applied tasks and can include the classification of the components and areas of the earth surface on the basis of brightness, spectral and spatial features and distinctions. In some cases we need image correction and their "matching" with mapping projections, while in other cases we need their "vectorization" - translation into thematical digital maps which are passed on to what we call "geodata" systems. Finally, for the identification of temporal image variations it is necessary to compare all the incoming data with those gathered before. All of these things, of course, call for the use of high-performance specialized computer complexes.
As was mentioned before, the large volumes of digital data associated with the transmitted images pose a range of serious problems before the designers of such new equipment. The requirements of rapid data transmission along telecommunications networks, their rapid processing and registration, come into conflict with the technical parameters of the available equipment. First of all, the passage of the transmitting channels is limited. And even the most advanced computer devices are lacking the required rate of response and the memory capacity. Of great importance in this situation is a special mode of image processing - what we call data compression.
This technique can be implemented in many ways, but all of them are characterized by what we call the "compression ratio" or the ratio between the volumes of data input and output. And the higher this index, the more effective is the method employed. At times it may be possible to "compress" the initial volume of data by 10 to 50 times.
However, the growing compression factor is restricted in principle by several objective factors, such as the level of distortions introduced into the image at its compression. That is why the choice of the permissible level of distortions is important. Let us say, in the processing of "visual" images the peculiarities of their visual perception are taken into account (in this case the distortions must be hardly visible to the viewer), but in other cases this criterion cannot be applied. For example, in remote sensing of the
earth surface the "informative value" of the picture is so great that it puts the highest requirements on the exact registration of every spot of the brightness field. The fixation of such images is usually conducted for the subsequent multipurpose utilization. This is done under stringest control and with the use of methods causing little or no distortion, such as differential coding, coding with transformation; interpolation modes based upon comprehensive segmentation. Each of these has advantages and disadvantages of its own which makes studies in this field important to this day.
Similar problems are being dealt with at the RAS Institute of Image-Processing Systems, and considerable progress has already been achieved in this area. It has been possible to work out algorithms and methods of compression or decompression of images which are far superior to the existing ones. They ensure an increased compression factor, make it possible to control images according to the most stringent criterion of maximum error of reproduction of every spot (pixel) of digital image, and permit what we call "multiresolution" or rapid restoration of reduced copies, etc. This kind of data processing is used in many applied studies. Based upon it
is digital television and computerized polygraphic systems, image databases (space photos, photos of works of art, trademarks, etc.) with remote access through the Internet, etc.
Computer image processing also covers a very different and rather unexpected area of modern research - the area of medico-biological investigations and diagnostics. To be more exact, this includes X-ray photos (of lungs, joints and other internal organs), tomographic data (ultrasonic and X-ray images obtained by the method of magnetic nuclear resonance, etc.), studies of tissue and blood preparations and also of the retina of the eye (for iridodiagnostics) and many other applications. Even purely "unidimensional" diagnostic materials, like cardiograms and encephalograms, can be input into a computer and processed as images because their analysis and interpretation by medical experts in this case can lead to more accurate diagnosis. Having said that, let us consider several concrete examples of using such systems.
As was mentioned before, one of the more developed areas of such studies is associated with iridodiagnostics. The thing is that the eye-ground is the only "area" wherein the system of blood vessels is open to direct and non-invasive observations. The earliest symptoms of its lesions in cases of diabetes, arterial hypertension or atherosclerosis are the changing ratios of the diameter of arteries and veins, local "jumps" in the diameter of blood vessels, their increased convolutions, etc. This underlines the importance of developing methods of reliable assessment of pathological changes in the blood vessels of the iris. But experts working on such systems run into a range of problems. They need, for example, digital images of high resolution and high accuracy for measuring vessels' diameter because this makes it possible to assess what we call the local haemodynamics of the iris.
Experts of the RAS Institute of Image-Processing Systems have developed an original set of instruments for studies of the iris which makes it possible to arrive at a "quantitative" assessment of its pathologies. The system first traces (determines the median line of every vessel) the whole vascular system on the basis of which doctors then determine the geometry of the veins and arteries. This includes the length of a blood vessel and the parameters describing every section of it. Coordinates are determined of the point of scanning of the vessel at every step of its tracing; the local width of a vessel; the direction to the next tracing point and the number of vessel branches at a given point. The sum total of these data makes it possible to diagnose the condition of the whole iris and the eye-ground using 11 parameters. This provides the foundation of a database which also includes what we call general clinical information about the patients. At the final stage of the work of this system a program is activated of automatic assessment of the degree of pathology on the basis of expert conclusions.
This unique analytical complex can also be used for general therapeutic applications and in obstetric practice for early diagnostics and effective treatment of vascular disorders. As different from European analogues, this equipment also makes it possible to analyze sub-clinical morphological changes.
Another area of this research is associated with the digital analysis of crystallograms of lacrimal fluid (serves as an indicator of exchange upsets in different pathologies). From the biochemical standpoint, they represent a multicomponent system the studies of which by traditional methods require large amounts of the liquid itself and expensive laboratory equipment and reagents. And even with all that it is impossible to carry out such analyses for the presence of all the lacrimal components.
In view of the above, a new diagnostics system was proposed based upon the crystallographic method (providing a picture of the fundamental structure of the substances involved). Researchers of the Ophthalmology Chair of the Volgograd Medical Academy conducted an analysis of crystallogram images, which had been previously divided by experts into normal and pathological ones, and identified what they call "global" diagnostic symptoms helping in such classification (in the absence of eye disorders the lacrimal crystal should be transparent, contain long rays of predominantly one orientation, having clear borders and proceeding from a common center; in other cases the picture is different which is a sign of some pathology).
Today experts in most, if not all, countries are working on problems of image processing. These studies are part and parcel of the information science and rest upon such disciplines as mathematics, physics and biology, to name but a few. In the view of Prof. Soifer, the deeper we investigate what he calls the information nature of images, the broader field of research activities opens up before us today.
V. Soifer, "Computer-Image Processing", Vestnik RAN, Vol. 71, No. 2, 2001
Опубликовано 14 сентября 2018 года
Картинка к публикации:
Полная версия публикации №1536956826 →
Главная → КОМПЬЮТЕРНЫЕ ТЕХНОЛОГИИ → IMAGE-PROCESSING POTENTIAL
При перепечатке индексируемая активная ссылка на PORTALUS.RU обязательна!