Abstract
Image fusion involves combining multiple images of differing modalities (e.g. infrared and visible light radiation) to create a composite that contains complementary information from the inputs. Most methods of fused image assessment have involved some kind of subjective quality rating, which is then correlated with computational metrics. A different approach recently adopted (Dixon et. al, 2005; submitted) is to apply a task to the assessment process. Extending a 2AFC paradigm previously used, the current paper compares how quickly participants stated whether a soldier was left or right of a clearing of trees in frames of a fused image sequence. False colored fused images were used, created by applying the infrared image to a red color plane, and the visible light image to a green color plane (1 condition), as well as adding a uniform blue additional color plane (1 condition), and comparing these with the monochrome Dual-Tree Complex Wavelet Transform (DT-CWT) method. In addition, two levels of JPEG2000 compression were applied, as well as an uncompressed condition. Participants were also asked to make subjective quality assessments of the images. The results showed significantly faster reaction times for the two color-fusion methods over the DT-CWT method, although no significant effect of compression was found. The subjective quality ratings showed main effects of both fusion method and compression, and an interaction. The results suggest that application of color to a given fused image assessment task can be beneficial, whilst subjective quality ratings should be used with care.
This work has been partially funded by the UK MOD Data and Information Fusion Defence Technology Centre. The original ‘UN Camp’ infrared and visible images were kindly supplied by Alexander Toet of the TNO Human Factors Research Institute. These images are available online at
www.imagefusion.org.