Deconvolution of MODIS imagery using multiscale maximum entropy

Abstract

A multiscale maximum entropy method (MEM) for image deconvolution is implemented and applied to MODIS (moderate resolution imaging spectroradiometer) data to remove instrument point-spread function (PSF) effects. The implementation utilizes three efficient computational methods: a fast Fourier transform convolution, a wavelet image decomposition and an algorithm for gradient method step-size estimation that together enable rapid image deconvolution. Multiscale entropy uses wavelet transforms to implicitly include an image's two-dimensional structural information into the algorithm's entropy calculation. An evaluation using synthetic data shows that the deconvolution algorithm reduces the maximum individual pixel error from 90.01 to 0.34%. Deconvolution of MODIS data is shown to resolve significant features and is most effective in regions where there are large changes in radiance such as coastal zones or contrasting land covers.

Acknowledgements

We would like to thank Jack Xiong, Brian Wenny and Gerhard Meister for contributing MODIS Aqua characterization models. Thanks also go to Thomas Shroeder, Young Je Park and Edward King for reviewing this letter. This work was carried out under the funding of the CSIRO Internship program and Peter Turner's contribution was funded by the CSIRO Wealth from Oceans Flagship.