Study On Image Correction Algorithm Based On Stationary Random Field

The photogrammetric processing of remote sensing image has been becoming more and more important in contemporary photogrammetry. There have been various kinds of methods and relevant practical softwares for remote sensing image processing. Geometric correction for which there have been various processing methods, is one of the important parts of remote sensing image processing. In practice, the traditional model based on polynomial functions and the model based on collinearity equation are representative and used commonly. However both correction models have respective advantages and disadvantages.Random Fields theory begins in the early of 20th century .It studies the statistical rules of dynamic and whole random phenomena with the spatial variables change with the help of the principles of probability theory and statistics. Kriging Algorithm is an interpolation algorithm based on smooth random fields and intrinsic random fields, which can get perfect interpolation precision and is widely used in the field of the Geographic Information System. Recently, Kriging Algorithm has also been applied in the field of image process and this article attempts to apply it to rigorous correction for remote sensing image. Recent research demonstrates that terrain change is essentially random and it can be modeled as a random field. The distortions caused by terrain change distribute randomly in image, therefore the pixel's errors of the rough post-correction image along x-axis and y-axis can be taken as a separate random field and the random field can be fitted and stimulated. The semivariogram of every pixel's errors along the north and east direction in image space can be obtained by the point errors of control points after rough geometric correction. The semivariance can be obtained from the semivariogram and taken as a correlation measurement in this method. Every control point errors influence weight to every pixels errors can be obtained through the semivariogram, then every pixel's errors along x-axis and y-axis can be got .The pixel's errors are corrected in the final results andthe final correction precision improves.Although the accuracy is very close in the plane areas compared with other traditional algorithms according to the results of experiments, Kriging Algorithm applied in geometric correction performs better and attains comparatively high accuracy in rolling areas. And it also provides a good method for geometric correction, especially for the rolling areas without DEM. Besides, the image is classified according to the DEM and vigorous geometric correction for rolling areas can perform locally. Therefore it can be taken as an good attempt in practice on vigorous geometric correction for rolling areas by the method this paper discusses and provides.