Abstract
This paper addresses the issue of missing data reconstruction for partially sampled, two-dimensional, rectangular grid images of differentiable random fields. We introduce a stochastic gradient–curvature (GC) reconstruction method, which is based on the concept of a random field model defined by means of local interactions (constraints). The GC reconstruction method aims to match the gradient and curvature constraints for the entire grid with those of the sample using conditional Monte Carlo simulations that honor the sample values. The GC reconstruction method does not assume a parametric form for the underlying probability distribution of the data. It is also computationally efficient and requires minimal user input, properties that make it suitable for automated processing of large data sets (e.g. remotely sensed images). The GC reconstruction performance is compared with established classification and interpolation methods for both synthetic and real world data. The impact of various factors such as domain size, degree of thinning, discretization, initialization, correlation properties, and noise on GC reconstruction performance are investigated by means of simulated random field realizations. An assessment of GC reconstruction performance on real data is conducted by removing randomly selected and contiguous groups of points from satellite rainfall data and an image of the lunar surface.
Similar content being viewed by others
Notes
Ergodicity implies that the spatial average of an observable is equal to the respective ensemble average. In contrast, we assume that the spatial average over the entire grid equals the sample average. For example, for a grid with length comparable to the random field correlation length, ergodicity is not satisfied. Nevertheless, the gradient and curvature constraints over the grid have definite values (not necessarily identical to the ensemble values); our assumption is that the sample drawn from the grid gives accurate estimates of the constraints over the grid.
By interacting grid nodes we imply the pairs and triplets of adjacent nodes that contribute to the gradient and curvature constraints given by (2) and (3), respectively.
A circular stencil would correspond to the Euclidean distance based KNN classification algorithm, k being the number of sampling points inside the stencil.
Other definitions of misclassification rate can also be used based on smoother distance functions between class values.
References
Acker JG, Leptoukh G (2007) Online analysis enhances use of NASA Earth Science data. EOS, Trans Am Geophys Union 88(2):14, 17
Aizerman M, Braverman EM, Rozenoer LI (1964) Theoretical foundations of the potential function method in pattern recognition learning. Autom Remote Control 25:821
Atkinson PM, Lewis P (1999) Advances in remote sensing and GIS analysis. Wiley, New York
Briggs IC (1974) Machine contouring using minimum curvature. Geophysics 39:39–48
Cherkassky VS, Gehring D, Mulier F (1996) Comparison of addaptive methods for function estimation from samples. IEEE Trans Neural Networks 7:969–984
Chica-Olmo M, Abarca-Hernández F (2000) Computing geostatistical image texture for remotely sensed data classification. Comput Geosci 26:373–383
Cressie N (1993) Spatial statistics. Wiley, New York
Dasarathy BV (ed) (1991) Nearest neighbor norms: NN pattern classification techniques. IEEE Computer Society Press, Los Alamitos
Diggle PJ, Ribeiro PJ (2007) Model-based geostatistics. Springer series in statistics, Springer, New York
Elogne SN, Hristopulos DT (2008) Geostatistical applications of Spartan spatial random fields. In: Soares A, Pereira MJ, Dimitrakopoulos R (eds) GeoENV VI geostatistics for environmental applications, vol 15 of series quantitative geology and geostatistics, Springer, Berlin, pp 477–488
Elogne SN, Hristopulos D, Varouchakis E (2008) An application of Spartan spatial random fields in environmental mapping: focus on automatic mapping capabilities. Stoch Env Res Risk Assess 22:633–646
Feuer A, Goodwin G (2005) Reconstruction of multidimensional bandlimited signals from nonuniform and generalized samples. IEEE Trans Signal Process 53:4273–4282
Forbes F, Fort G (2007) Combining monte carlo and mean-field-like methods for inference in hidden Markov random fields. IEEE Trans Image Process 16:824–837
Foster MP, Evans AN (2008) An evaluation of interpolation techniques for reconstructing ionospheric TEC maps. IEEE Trans Geosci Remote Sens 46:2153–2164
Hansen TM (2007) mGstat: a geostatistical Matlab® toolbox
Hristopulos DT (2003) Spartan Gibbs random field models for geostatistical applications. SIAM J Sci Comput 24:2125–2162
Hristopulos DT, Elogne S (2007) Analytic properties and covariance functions of a new class of generalized Gibbs random fields. IEEE Trans Inf Theory 53:4667–4679
Hristopulos DT, Elogne SN (2009) Computationally efficient spatial interpolators based on spartan spatial random fields. IEEE Trans Signal Process 57:3475–3487
Hsu CW, Lin CJ (2002) A comparison of methods for multi-class support vector machines. IEEE Trans Neural Networks 13:415–425
Kaheil YH, Gill MK, McKee M, Bastidas LA, Rosero E (2008) Downscaling and assimilation of surface soil moisture using ground truth measurements. IEEE Trans Geosci Remote Sens 46:137–1384
Kanevski M, Maignan M (2004) Analysis and modelling of spatial environmental data. EPFL Press, Lausanne
Keller JM, Gray MR, Givens JAJ (1985) Fuzzy k-nearest neighbor algorithm. IEEE Trans Syst, Man, Cybern 15:580–585
Kitanidis PK (1997) Introduction to geostatistics: applications to hydrogeology. Cambridge University Press, New York
Kokaram A (2004) On missing data treatment for degraded video and film archives: a survey and a new Bayesian approach. IEEE Trans Image Process 13:397–415
Kokaram A, Godsill S (2002) MCMC for joint noise reduction and missing data treatment in degraded video. IEEE Trans Signal Process 50:189–205
Kramer PR, Kurbanmuradov O, Sabelfeld K (2007) Comparative analysis of multiscale Gaussian random field simulation algorithms. J Comput Phys 226:897–924
Larsson E, Stoica P, Li J (2002) Amplitude spectrum estimation for two-dimensional gapped data. IEEE Trans Signal Process 50:1343–1354
Matheron G (1971) The theory of regionalized variables and its application. Les Cahiers du Centre de Morphologie Mathematique de Fontainebleau
Papadimitriou CH, Steiglitz K (1982) Combinatorial optimization: algorithms and complexity. Prentice Hall, Upper Saddle River
Rohde GK, Aldroubi A, Healy DM (2009) Interpolation artifacts in sub-pixel image registration. IEEE Trans Image Process 18:333–345
Rossi RE, Dungan JL, Beck LR (1994) Kriging in the shadows: geostatistical interpolation for remote sensing. Remote Sens Environ 49:32–40
Sanchez-Beato A, Pajares G (2008) Noniterative interpolation-based super-resolution minimizing aliasing in the reconstructed image. IEEE Trans Image Process 17:1817–1826
Sandwell DT (1987) Interpolation of GEOS-3 and SEASAT altimeter data. Geophys Res Lett 14:139–142
Shi J, Reichenbach S (2006) Image interpolation by two-dimensional parametric cubic convolution. IEEE Trans Image Process 15:1857–1870
Stein ML (1999) Interpolation of spatial data: some theory for kriging. Springer, New York
Vapnik V, Lerner A (1963) Pattern recognition using generalized portrait method. Autom Remote Control 24:774–780
Vapnik V (1998) Statistical learning theory. Wiley, New York
Wackernagel H (2003) Multivariate geostatistics. Springer, Berlin
Weber A (2004) USC-SIPI image database, Vol 3. University of Southern California, Los Angeles
Yaglom AM (1987) Correlation theory of stationary and related random functions I. Springer, New York
Zhang X, Wu X (2008) Image interpolation by adaptive 2-d autoregressive modeling and soft-decision estimation. IEEE Trans Inf Theory 17:887–896
Žukovič M, Hristopulos DT (2009a) classification of missing values in spatial data using spin models. Phys Rev E 80: Art. no. 011116
Žukovič M, Hristopulos DT (2009b) Multilevel discretized random field models with “spin” correlations for the simulation of environmental spatial data. J Stat Mech: Theory Exp. Art. no. P02023
Acknowledgments
This research project has been supported in part by a Marie Curie Transfer of Knowledge Fellowship of the European Community’s Sixth Framework Programme under contract number MTKD-CT-2004-014135. We also acknowledge support for a short visit by Prof. Žukovič at TUC that helped to finalize this paper from the Hellenic Ministry of Education—Department of Inter-University Relations, the State Scholarships Foundation of Greece and the Slovak Republic’s Ministry of Education through the Bilateral Programme of Educational Exchanges between Greece and Slovakia. We are grateful to Prof. M. Kanevski (Université de Lausanne, Switzerland) for providing us with the full version of the GeoSVM software. Rainfall data used in this study were produced with the Giovanni online data system, developed and maintained by the NASA Goddard Earth Sciences (GES) Data and Information Services Center (DISC). We also acknowledge the TRMM mission scientists and associated NASA personnel for the production of the rainfall data used in this research effort.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Žukovič, M., Hristopulos, D.T. Reconstruction of missing data in remote sensing images using conditional stochastic optimization with global geometric constraints. Stoch Environ Res Risk Assess 27, 785–806 (2013). https://doi.org/10.1007/s00477-012-0618-5
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00477-012-0618-5