K.K. Vasiliev - Dr. Sci. (Eng.), professor, Ulyanovsk State Technical University. E-mail: vkk@ulstu.ru
V.E. Dementiev - Ph.D. (Eng.), associate professor, Ulyanovsk State Technical University. E-mail: vitawed@mail.ru
N.A. Andriyanov - post-graduate student, Ulyanovsk State Technical University. E-mail: nikita-and-nov@mail.ru

Recently special urgency of the problem associated with the development and research of image processing algorithms and video sequences in a variety of machine vision systems . This is explained by the ever-increasing volume of storing and processing of digital images and the increasing ability of modern computing. A special role is played reconstruction tasks or image evaluation at a given point or area. Effective solution to this problem can be considered as pre-processing for filtering, image segmentation , detection of anomalies on them of various kinds. There are several methods that allow you to restore the image. It was described how to restore an image using singular value decomposition of matrices with gaps; and the recovery method using Kohonen self-organizing maps was considered. However, these methods are characterized by relatively large errors and difficulties related to the practical implementation. In this paper we define the parameters of existing image approximating model and perform on these parameters needed to forecast point and the image area. A comparative analysis of the Kalman filter and the Wiener filter in the evaluation of doubly stochastic random fields is considered in the text, as well as a comparison parameter identification algorithms based on the Kalman filter and using a sliding window. Analysis of the variance showed the possibility of filtering errors of estimation of the model parameters using nonlinear recursive filtering algorithms. In most cases for image restoration by the parameters most appropriate to use vector Kalman filters.