Идентификация телевизионных изображений системы технического зрения на основе математического аппарата кубических нормализованных B-сплайнов

Introduction. The solution to the problem of television images identification under the creation of autonomous robots, vision systems, and noisy image analysis systems is considered. The question is, for example, on severe observing conditions hindering the registration process, and aprior information on the type of background noise. The work objective is to develop and evaluate the efficiency of the method for image edge detection (two-dimensional signal) against the background of pulse noise using the mathematical apparatus of cubic B-splines. Materials and Methods. Involving intense background noise, spline-approximation of discrete values of signals and images is usually unproductive and leads to raw errors. In this case, the method of differentiating the image line against the noise background allows calculating the signal derivative with sufficient accuracy. Taking into account the information on the behavior of the first derivative, local maxima in the image line against the noise background are defined. The task of television image edge detection is solved by a new technique of spline-differentiation. For this, the image matrix is divided into lines and columns; the differentiation is performed; and then the edge extraction operators are calculated. Unlike the known approaches, the differentiation takes into account data on the intensity in the whole image line. This minimizes the noise effect. Image edges are defined using an intensity gradient. The resulting spline-differentiation algorithm is used for mathematical modeling. Research Results. The authors of the paper for the first time propose a high-precision method of digital differentiation of two-dimensional signals. This approach allows calculating values of the two-dimensional signal derivative and its gradient with sufficiently high accuracy. With that, there is no need to use standard numerical differentiation procedures which are incorrect in themselves. Lena test image distorted by pulse noises of “dead pixels” and “salt-pepper” is processed by the Sobel operator and the spline-differentiation method. Values of еско, SNR and SNRF are tabulated and analyzed. For the Lena test image, the gain in decibels was as follows: according to the MSD (mean-square deviation) - 1.6 ÷ 2.7; relative to peak signal/ SNR noise ratio - 8 ÷ 9.4; relative to peak signal/ MSD noise of SNRF background - 11 ÷ 12. Discussion and Conclusions. Under the conditions of rapid development of microtechnology, the problems solved with the help of vision systems take a new way of application. This proves the relevance of research in the field of increasing the efficiency and stability of methods and algorithms for digital processing of two-dimensional signals. The experiments show that the presented technique has considerably higher noise immunity than algorithms based on standard differentiation procedures.