Fuzzy Parameters Integrated Markov Random Field (MRF) Model for Super Resolution Mapping (SRM) Over Vague Land Cover Regions

The main objective of this study is to improve the Markov Random Field (MRF) based Super Resolution land cover Mapping (SRM) technique to optimize it performance for vague land cover classes. Additionally, the method could be used in order to minimize the point spread function (PSF) effects in multispectral data. In this regard, the study proposes a fuzzy class parameter estimated MRF model for the SRM generation. The paper describes two approaches for the pixel membership determination: spectral angle (SA) of each pixel and the fuzzy c-means (FCM) algorithm. Fuzzy class mean and the covariance measurements were derived using these memberships to model the class distributions within the MRF model. The use of fuzzy memberships can improve the overall posterior energy functions to reach a minimum, by taking both contextual and likelihood estimations of the V-I-S classes. The technique was tested using a WORLDVIEW-2 satellite image, acquired over a semi urban area in Colombo, Sri Lanka. Based on Visual interpretation three major land-cover types; vegetation, soil and exposed grass, and impervious surface (V-I-S) with low medium and high albedo were selected for the study. The bench mark reference data was generated using Maximum Likelihood Classification (MLC) performed on the same data resampled to 1 m resolution. The scale factor was set to be (S) =2, to generate SRM of 1 m resolution. The smoothening parameter (λ) which balances the prior and likelihood energy terms were tested in the range from 0.3 to 0.9. Fuzzy and conventional SRM were generated separately for each method. Both the methods produce satisfactory results with respect to its conventional counterpart with their own advantageous. Finally, the comparison of the model with the two non-parametric approaches: Support Vector Machine (SVM) and Neural Network (NN) are reported.