On the use of joint sparse representation for image fusion quality evaluation and analysis

In this paper, a Spare Representation (SR) based fusion quality evaluation and analysis method is proposed. This method employs Joint Sparse Representation (JSR) to extract the source image remnants after fusion. These atom-level remnants indicate the fusion quality intuitively, and permit the analysis of fusion effect in learned feature space. Our analysis results indicate that high salient atoms always present poor expressions in fusion results. An improved fusion rule is designed to emphasis high salient atoms accordingly. In experiments, the effectiveness of our method was verified and the characteristics of atom JSR remnants were investigated in detail first. Then the new fusion rule was tested to demonstrate the value of JSR remnant analysis. The objective and subjective comparison results indicate that the proposed analytical evaluation metric can measure fusion quality and analysis atom fusion effect accurately. The new fusion rule provides a valuable alternative for SR fusion algorithm design. (C) 2019 Elsevier Inc. All rights reserved.