Publishing house Radiotekhnika

"Publishing house Radiotekhnika":
scientific and technical literature.
Books and journals of publishing houses: IPRZHR, RS-PRESS, SCIENCE-PRESS

Тел.: +7 (495) 625-9241


Optimization of two images mutual information estimation


G.L. Safina – Ph.D.(Eng.), Associate Professor, Department «Applied Mathematics», National Research Moscow State University of Civil Engineering
A.G. Tashlinsky – Dr.Sc.(Eng.), Professor, Head of Department «Radio Engineering», Ulyanovsk State Technical University
M.G. Tsarev – Post-graduate Student, Ulyanovsk State Technical University

Shannon's mutual information (MI) is one of the measures of two images similarity. It is used for their mismatch. The MI estimate can be determined from a local sample of image samples. The volume of useful information contained in brightness differences depends on the distance between the samples that are in the local sample. For a small distance, the noise suppresses useful information. If the distance is too long, the useful information is suppressed by the random component of the brightness, that increases as the correlation coefficient between the samples decreases. Thus, there is an optimal distance between the samples, providing with a given sample size a maximum of information on the mismatch of the images. The problem to determine this distance under conditions of additive noisy images is solved.
The solution of the problem is complicated by the fact that it is necessary to use all the differences in the brightness of the samples from the local sample to determine the MI of images. The solution is based on the fact that the maximum information about mismatch of images occurs with the maximum ratio of the mathematical expectation module of the MI gradient to its standard deviation. The analysis of the obtained expression for the gradient of mutual information estimation is analyzed. It is invariant to the form of the image mismatch model. It is shown that the minimum possible size of a local sample is equal to three samples. The relations allowing to determine the optimal distance are obtained for this local sampling size. They are generalized to the case of an arbitrary sampling size.
The studies have shown that increasing the sample size slightly influences on the optimal distance, which, as a rule, is much smaller than the size of the studied images. The images are discrete, this fact limits the local sampling size. On the other hand, increasing the sampling size leads to increasing MI volume. This contradiction can be resolved by using a set of optimized samples in the MI calculation. In this case the samples are chosen in different domains of the image, for example, regularly or randomly. For a given size each sample has a maximum of information about images mismatch. The required information volume can be obtained by specifying the desired number of samples.

  1. Stanier S.A. Improved image-based deformation measurement for geotechnical applications Stanier // Canadian Geotechnical Journal. 2015. T. 53. № 5. P. 727−739.
  2. Shalev-Shwartz Shai and Zhang Tong. Accelerated proximal stochastic dual coordinate ascent for regularized loss minimization // Mathematical Programming. 2016. V. 155. № 1. P. 105−145.
  3. Rocco I., Arandjelovic R., Sivic J. Convolutional neural network architecture for geometric matching // Proc. CVPR. 2017. T. 2. P. 6148−6157.
  4. Tashlinskij A.G., Voronov S.V. Ispol’zovanie vzaimnoj informaczii kak czelevoj funkczii kachestva oczenivaniya parametrov izobrazhenij // Avtomatizacziya proczessov upravleniya. 2014. № 3 (37). S. 32−39.
  5. Tashlinskiy A.G., Zhukova A.V. Effectiveness of correlation and information measures for synthesis of recurrent algorithms for estimating spatial deformations of video sequences // CEUR Workshop Proceedings. 2017. V. 1901. P. 235−239.
  6. Shannon C.E., Weaver W. The mathematical theory of communication // The Mathematical Theory of Communication. 1949. P. 29−125.
  7. Maes F.,Vandermeulen D., Suetens P. Medical image registration using mutual information // Proc. IEEE. 2003. V. 91. № 10. P. 1699−1722.
  8. Tashlinskii A.G., Safina G.L., Voronov S.V. Pseudogradient optimization of objective function in estimation of geometric interframe image deformations // Pattern recognition and image analysis. 2012. V. 22. № 2. P. 386−392.
  9. Xoreva A.M., Tashlinskij A.G., Smirnov P.V. Vy’bor konechny’x raznostej pri naxozhdenii psevdogradienta czelevoj funkczii v proczedurax oczenivaniya mezhkadrovy’x deformaczij izobrazhenij // Radiotexnka. 2012. № 9. S. 56−60.
  10. Parzen E. On Estimation of a Probability Density Function and Mode // Annals of Math. Statistics. 1962. V. 33. P. 1065−1076.
  11. Tashlinskij A.G., Voronov S.V., Voronov I.V. Analiz czelevy’x funkczij v zadache oczenivaniya vzaimny’x geometricheskix deformaczij izobrazhenij // Avtomatizacziya proczessov upravleniya. 2013. № 4(34). S. 26−29.
  12. Viola P., Wells W.M. Alignment by maximization of mutual information // International Journal of Computer Vision. 1997. V. 24. P. 137−154.
June 24, 2020
May 29, 2020

© Издательство «РАДИОТЕХНИКА», 2004-2017            Тел.: (495) 625-9241                   Designed by [SWAP]Studio