Polar ice image segmentation using improved estimation and normalization of illumination

Cloud computing is a large-scale paradigm in computing that is driven by the economies of scale. It is a pool of the virtualized, abstracted, dynamically scalable, storage, platforms, managed power of computing and the services that are delivered based on demand to its external customers. With advent of cloud services, remote sensing data have been effectively used for identifying Region of Interest in various industries. Cloud computing will be well suited for all the computationally-intensive and also the data-intensive services of remote sensing. One popular application of remote sensing data is the estimation of ice sheet thickness in sub glacial topography and estimate mass balance of large bodies of ice. Segmentation plays a very important role in identifying the mass. The Automated polar ice-based image interpretation normally consists of the lower level segmentation based on a higher level of classification and on the basis of this homogeneity criteria or its definitions which is a region boundary, and this will be partitioned into the image pixels of a number of necessary regions. The statistical as well as the structural characteristics for such regions are used with the process of classification for deriving the nature and its class for all the regions and the success of its final interpretation in the polar ice will depend upon the performance of its low-level segmentation. This work proposes a novel technique for ice image segmentation using a modified estimation and normalization of illumination technique used in the Retinex algorithm along with Fuzzy C Means segmentation technique and Gaussian Mixture Model technique based on the layer of ice. Experiments carried out with the proposed technique shows improved accuracy in segmentation.