Abstract
Conference Title: 2016 IEEE 8th International Conference on Intelligent Systems (IS) Conference Start Date: 2016, Sept. 4 Conference End Date: 2016, Sept. 6 Conference Location: Sofia, Bulgaria Geostatistics and the theory of regionalized variables have been increasingly found useful in many applications of image processing. The semivariogram is the cornerstone of this spatial statistics, which can be implemented as an effective feature for pattern classification. This paper introduces a kriging-based distortion approach, and explores several well-known methods for pattern matching of the empirical semivariograms, including the spectral distortion measures, dynamic-time warping, and sample entropy. The findings provide insights into the utilization of several algorithms for the semivariogram-based pattern comparison under different settings. In particular, the usefulness of the proposed approach applies to situations where samples are limited, making hindrances for many pattern classifiers, which usually rely on some certain amount of sufficient training data.