A.G. Tashlinskii, A.M. Horeva, P.V. Smirnov
As a result for interframe geometric image distortions the same scene elements have different coordinates in different image frames, which can be represented as transformation of reference image coordinate system into coordinate system of distorted image. One of the most promising class for interframe geometrical distortions estimation is pseudogradient procedures class. Taking into account the discreet of digital images pseudogradient compupation using local sample and current distortions parameters estimations is lead to finite differences computation. The highest convergence speed of distortion parameters estimations can be obtained by using for local sampling the most informative image samples. It’s shown that for estimation of goal function derivatives there is optimal increment value for parameters and basic axes, which depends on parameters mismatch and image correlation properties. Performance of procedures can be improved by adapting the increment at each iteration. Optimal criterion for goal function pseudogradient calculating is the potential accuracy of the estimation parameters determined by the Cramer-Rao inequality. It is shown that the condition of maximum information can be reduced to the condition of maximizing the mean ratio of pseudogradient to its standard deviation. Using the optimized values of the increments let us significantly (several times) reduce the computational cost for the same estimation accuracy.