E.A. Samoylin - D.Phil, Professor of faculty, Military educational-scientific center of air forces «Military air academy professor N.E. Zhukovskiy and Y.A. Gagarin». E-mail: es977@mail.ru M.A. Pantuyhin - Post-graduate Student, Military educational-scientific center of air forces «Military air academy professor N.E. Zhukovskiy and Y.A. Gagarin». E-mail: ol-max@mail.ru

The algorithm for constructing a piecewise-linear approximation of the object-s outer contour on the distorted digital image using adaptive linear neuroelements based on the step-by-step tracing procedure is offered. Currently there is a wide range of tasks related to the patterns recognition. The edges detection and image segmentation stands the major place among them. The masks of Prewitt, Robinson, Kirsh and others are widely used to solve it. Their algorithm is based on an analysis of the brightness - difference on the image. Though, the images, which are taken by technical assets, contain noises and distortion. In practice, an adaptive noise and blur are present. They lead to false and pass the true contour elements. To form the object-s outer contour on the image is offered to use a neural network to generate estimates of the parameters of the linear parts make up the whole circuit. These statements form the following algorithm to construct the piecewise linear approximation of the object-s contour on the image. The first step of the algorithm uses the Canny-s method to build the binary contour preparation. The second step is tracing the binary contour preparation using the offered step-by-step procedure to find the ordered set of points defining the outer contour of the object. The third step is dividing these set into subsets to use them as training sets for neural network consists of a single neuroelement to find the parameters of linear dependencies that determine the behavior of the selected subsets. The final step is to close the linear segments by finding the intersection points. Using the offered algorithm under the influence of the distorted factors provides to form the thinnest object-s outer contour on the image with the less frequency of false contour elements and approximately constant frequency missing contour elements compared with the Canny-s method. The disadvantage of the offered algorithm is the higher computational costs, which could be minimized by paralleling the computational procedures. The resulting piecewise linear contour could be used for objects - recognition.

 

1. Djatlov E.I. Mashinnoe zrenie (analiticheskijj obzor) // Matrichnye sistemy i mashiny. 2013. № 2. S. 32-40.
2. Potapov A. Sistemy kompjuternogo zrenija: sovremennye zadachi i metody // ControlEngineeringRossija. 2014. № 1(49). S. 20-26.
3. Prokopenko V.T., Trofimov V.A., SHarok L.P. Psikhologija zritelnogo vosprijatija / Uchebnoe posobie. SPb: SPbGUITMO. 2006. 73 s.
4. Gonsales R., Vuds R. Cifrovaja obrabotka izobrazhenijj. M.: Tekhnosfera. 2005. 1072 s.
5. Gruzman I.S., Kirichuk V.S., Kosykh V.P., Peretjagin G.I., Spektor A.A. Cifrovaja obrabotka izobrazhenijj v informacionnykh sistemakh / Uchebnoe posobie. Novosibirsk: Izd-vo NGTU. 2000. 168 s.
6. Samojjlin E.A., CHelakhov V.M.Povyshenie tochnosti approksimacii nestacionarnykh dannykh linejjnojj nejjronnojj setju metodom kortezhirovanija obuchajushhego mnozhestva // Nejjrokompjutery: razrabotka, primenenie. 2011. № 7. S. 20-26.
7. Samojjlin E.A., Pantjukhin M.A.Metod razmnozhenija nejjrosetevykh ocenok pri opredelenii parametrov linejjnojj zavisimosti po dannym, soderzhashhim normalno-anomalnye pomekhi // Nejjrokompjutery: razrabotka i primenenie. 2014. № 1. S. 41-45.
8. Canny J. A Computational Approach to Edge Detection // IEEE Transactions on pattern analysis and machine intelligence. Vol. PAMI-8. 1986. № 6. P. 679-698.
9. Rudnev V.A. Primenenie mikrokontrollerov dlja realizacii nejjronnykh setejj / Vestnik UjuRGU. 2012. № 16. S. 181-183.