An approach to select the appropriate image fusion algorithm for night vision systems

For many years image fusion has been an important subject in the image processing community. The purpose of image fusion is taking over the relevant information from two or several images to construct one result image. In the past many fusion algorithms were developed and published. Some attempts were made to assess the results from several fusion algorithms automatically with the objective of gaining the best suited output for human observers. But it was shown, that such objective machine-assessment does not always correlate with the observer's subjective perception. In this paper a novel approach is presented, which selects the appropriate fusion algorithm to receive the best image enhancement results for human observers. Assessment of the fusion algorithms' results was done based on the local contrasts. Fusion algorithms are used on a representative data set covering different use cases and image contents. These fusion results of selected data are judged subjectively by some human observers. Then the assessment algorithm with the best fit to the visual perception is used to select the best fusion algorithm for comparable scenarios.