A.V. Zhukova - Post-graduate Student, Ulyanovsk State Technical University E-mail: donskaya.alena@mail.ru S.V. Voronov - Ph. D. (Eng.), Associate Professor, Department «Radio Engineering», Ulyanovsk State Technical University E-mail: s.voronov@ulstu.ru

One of the most challenging problems in the processing of digital images defined by discrete sampling grids is to establish the cor-respondence between the conjugate points of two or more images. Image registration techniques developed for a variety of restrictions on the initial data can be divided into operating in the frequency and spatial domains. The need for solving image registration problem occurs in processing data received from the various sensors, computer vision, remote sensing of the Earth, identification of biometric parameters, robotics, medicine, national security. Under the conditions of a priori uncertainty recursive parameter estimation procedure employed in the spatial domain are more efficient and flexible often. They are usually based on the optimization of a multidimensional objective function characterizing the similarity measure (or difference) between pairs of images. There are many measures which can be used as objective functions. In conditions of strong brightness distortion of images, especially nonlinear, promising measures are those based on the information-theoretical approach. The most interesting among them in terms of the ratio of efficiency and computational complexity are Shannon and Tsallis mutual information, F information measure, excluding F information and energy of joint PDF. For each specific application the chose of the objective function is dependent on the characteristics of the image, type of deformation and the requirements of the problem being solved. The selected objective function largely determines the potential effectiveness of the synthesized recurrent image registration procedure. However, the issue of effec-tiveness criteria that allow evaluating the potential effectiveness of its use in the procedures using a priori form of the objective function are poorly studied. In this paper the mentioned above measures are compared via three criteria: maximal slope of the objective function on the examined range of registration parameters, effective range and slope growth area. Experimental studies were carried out on simulated images with the correlation function and the probability density function of brightness close to Gaussian. As an interfering factor an unbiased additive Gaussian noise is used. Experiments show that for the recursive estimation of parameters of image registration I measure of F information provides potentially greater rate of convergence of the estimated parameters. The same measure, due to the narrower width of the zone of inflection, provides a potentially less estimation error variance. According to these indicators slightly behind M measure of F information. Mutual information behind measures of F information in all investigated indices, and have a strong dependence on the noise level. In addition, it is worth noting that in terms of computational cost, measures of F information are far ahead of other measures, therefore, is often preferred.

 

1. Brown L.G. A survey of image registration techniques // ACM Computing surveys. 1992. V. 24. P. 325−376.
2. De Castro E., Morandi C. Registration of translated and rotated images using finite Fourier transform // IEEE Transactions on Pattern Analysis and Machine Intelligence. 1987. V. 9. № 5. P. 700−703.
3. Gonzalez R.C., Woods R.E. Digital image processing. Prentice Hall. New Jersey. 2002. 793 p.
4. Goshtasby A.A. Image registration. Principles, tools and methods: Advances in Computer Vision and Pattern Recognition. Springer. 2012. 441 p.
5. D\'Agostino E., Maes F., Vandermeulen D., Suetens P. An information theoretic approach for non-rigid image registration using voxel class probabilities // Med Image Anal. 2006. V. 6(3). P. 413−431.
6. Voronov S.V., Voronov I.V. Analiz ehffektivnosti informacionnykh mer kachestva privjazki izobrazhenijj // Radiotekhnika. 2015. № 6. S. 32−36.
7. Voronov S.V. Ispolzovanie vzaimnojj informacii kak celevojj funkcii kachestva ocenivanija parametrov izobrazhenijj // Radiotekhnika. 2014. № 7. S. 88−94.
8. Sevim Y., Atasoy A. Performance comparison of new nonparametric independent component analysis algorithm for different entropic indexes // Turkish Journal of Electrical Engineering & Computer Sciences. 2012. V. 20. P. 287−297.
9. Krasheninnikov V.R. Osnovy teorii obrabotki izobrazhenijj: uchebnoe posobie. Uljanovsk: UlGTU. 2003. 152 s.
10. Tashlinskijj A.G., Tikhonov V.O. Metodika analiza pogreshnostejj psevdogradientnogo izmerenija parametrov mnogomernykh processov // Izvestija VUZov. Ser. «Radioehlektronika». 2001. T. 44. № 9. S. 75−80.
11. Tashlinskijj A.G. Ocenivanie parametrov prostranstvennykh deformacijj posledovatelnostejj izobrazhenijj. Uljanovsk: UlGTU. 2000. 131 s.
12. Tashlinskii A.G. Pseudogradient Estimation of Digital Images Interframe Geometrical Deformations // Vision Systems: Segmentation & Pattern Recognition. Vienna (Austria): I‑Tech Education and Publishing. 2007. P. 465−494.