Nowadays all known methods and technical tools of image processing are adjusted to a small group of application tasks. They are quite reliable in case of minimal amount of unprompted disturbances (in case of jamming reliability of their functioning is doubtful). In that case the time of processing and decision making (of the class of complex images) is often longer than required (the system of search, detecting, observation and controlling – monitoring etc.).
A standard scheme of information transformations while solving the tasks of recognition is shown in pic.1. The stages of segmentation and tag formation have no theoretical solution nowadays (as especially in case of unprompted disturbances they are inconsistent, that is unable to give any definite answer). It's impossible to talk about any optimal solution (especially in real-time mode) to the problem of image classification without solving the mentioned problems with these two stages. There are lots of examples illustrating it: absence of reliable algorithms of personal identification with the help of a photo; absence of reliable algorithms of analyzing of TV images in monitoring systems for custom, security services and others; absence of reliable algorithms of not only analyzing of prescripts, but also texts with state standard fonts.
It should be noted that the basic calculating procedure of modern systems of image processing is folding, which is usually applied at the first two stages of recognition.
In contrast to the approach described above, human visual analyzer has a range of objective characteristics which are defined by innate mechanisms. These characteristics successfully solve the problem of reliable and almost immediate (in real time mode) recognition of any image, despite any disturbances (even if the disturbance hides the greater part of the object). The main objective characteristics are the following:
- integrity of perception — the process of image analyzing is realized as stable (inseparable) systematic whole;
- immediate perception — the process of recognition is realized at once, in large blocks and quickly, using a limited number of indicators;
- image handling — we use figurative descriptions, based on a subset of connected informational elements of an image, not an array of structurally unconnected figures.
We have developed theoretical apparatus, that designs inborn mechanisms of visual perception. TAPe-technology of random image processing was developed on their basis. The scheme of information transformations is shown in Pic.2.
The main content of the stage where the initial description is generated — is the processing of any image immediately and integrally. This principal and single procedure doesn't involve folding. The main content of the stage where the initial description is generated – is the processing of any image immediately and integrally. This principal and single procedure doesn't involve folding. The result is that we become absolutely independent from disturbances. They become part of the image and in some cases help to identify informative elements of the image. All possible variability of images is identically represented in the final variability of the image. So it becomes possible to process it in real-time mode with the help of conventional computing machinery. For example, lets process images with the limits of 1024×1024 with the level of brighteness gradation 256 (a standard processing task). Then the capacity of image variety — is a cosmic number equal to 2561048576. As the result of processing (according to our method) the capacity of halftone image variety will become less and will be represented with the help of relatively small number of tags (indicators), containing information about the most informative structural elements of the image. These elements are composite elements of the image.
The main target of the "analysis' stage is to process the results of the previous stage. The only procedure here is spatial differentiation, definition of structural (and compositional) links between the determined structural elements of the image.
On the third stage we generate graphic description in the form of frames, integral structural bonds on the image.
The stage of decision making involves comparison of graphic image with the sample, if we have some in the source, or formation of a sample if there is none.
When we avoid the stage of folding, algorithms of recognition (and comprehension) of any image is realized quicker and not less than 100 times more effective that the alternating definition systems, that use «roughly-precise» algorithms, which, in their turn are 106 times more effective than standard image recognition systems (as well as the systems based on matrix machines). Basic properties of TAPe-technology:
- pyramid-shape (allows to get multy-level description of an image «from top to the ground», both at the stages of analysis and decision making);
- universality (allows to recognize objects of any type in any frequency range);
- one-stage decision making (decisions are made at once and based on a limited number of reasons);
- compact size of description;
- invariance with respect to size changes, shifts and rotation limited by «physiological» angle of image perception;
- possibility to operate with «disturbed» images;
- reliability of recognition.