Analysis of Image Quality for Image Fusion via Monotonic Correlation

Abstract

This paper introduces a nonlinear correlation coefficient that exploits isotonic (or monotonic) regression. We refer to this correlation coefficient as the monotonic correlation (MC). This paper demonstrates how the MC scores the consistency between possible image quality (IQ) features and actual human performance, which is measured by a perception study. This paper also shows the relationship between the MC and the generalized likelihood ratio (GLR) test for the H 1 hypothesis that the IQ features are monotonically related to intrinsic human performance versus the null hypothesis that the relationship is arbitrary. Finally, the paper introduces a normalized GLR in order to assess the statistical significance of a high MC value. Using actual results from human perception experiments and the corresponding proposed IQ feature values for the imagery, the paper demonstrates how MC can identify worthy features that could be overlooked by traditional correlation values. The focus of the experiments center around the evaluation of IQ measures for image fusion applications.