UBC Theses and Dissertations
Analysis of Tsallis entropy-based measures for image fusion quality assessment and a new performance measure based on mutual information Sholehkerdar, Araz
Along with the significant increase in the number of image fusion methods, the research effort to develop effective image fusion quality assessment techniques have accelerated. In this Thesis, we focus on image fusion performance evaluation. This work is divided into three parts. First, the theoretical analysis over a proposed formalism for mutual information based on Tsallis entropy is conducted. The results show that the proposed formalism can not fulfill the desired behaviors that are expected from a mutual information-based measure. Second, two Tsallis entropy-based measures are employed for image fusion quality assessment. The closed-form expressions for the two measures are derived using an image formation model and considering weighted averaging as image fusion algorithm. The behavior analysis of the two measures is then performed regarding the important parameters of closed-form expressions including noise and entropy order. At the end, the utility of theoretical analysis is shown using experiments on real images. Third, a non-reference image fusion performance measure is introduced which utilizes mutual information based on Shannon’s entropy to extract the similarity between source and fused images. Here, a framework is designed that uses localized edge information in images to construct probability distributions. Estimating the joint probability distributions between images from the marginal distributions is also performed in this part. Finally, in order to to show the performance of proposed measure, the fusion quality of an image dataset is evaluated by our method and the outcome is compared with several available well-known image fusion quality measures.
Item Citations and Data
Attribution-NonCommercial-NoDerivatives 4.0 International