Establishment of Human Performance Baseline for Image Fusion Algorithms in the LWIR and SWIR Spectra

The performance benchmark process for discriminating between image fusion algorithms presented by Howell [3] and again by Moyer and Howell [6] establishes a performance goal for any aggregate function merging the source band imagery under investigation. It was reported by Moyer and Howell, in “Establishment of Human Performance Baseline for Image Fusion Algorithms in the LWIR and MWIR Spectra,” that increased P(ID) performances could be realized dependent upon how the source band information was presented. In that work, human observers were asked to identify displayed targets using a standard set of tracked military vehicles. The observers performed the ID tasks using LW and MW source band imagery concatenated on a single monitor, presented side by side on a single monitor, temporally interlaced source band imagery on a single monitor, and source band imagery presented to each eye of the observer in parallel. The research presented in this paper explored the impact dissimilar source band information had on the observer’s ability to identify targets without the aid of image fusion algorithms. The spectral bands under consideration for this effort were the LW and SW spectral bands. It was hypothesized that the different display techniques using dissimilar source spectral band information would better allow the observers to choose the portions of the source band images they needed to perform the visual discrimination task of identification above that achievable using the fused superposition images. It was determined after comparing the observers P(ID) performances using the superposition fused images to the resultant P(ID)s’ using these display techniques, that the performances using the superposition fused images were well below that which was achieved by the observers benchmark source band performance. Allowing the observer to view spectral source band imagery in different display formats without the aid of image fusion algorithms, which we refer to as “self-fusion,” allows the experimenter to establish an absolute benchmark for discriminating between image fusion algorithm performances. It was our intent to perform a mirror analysis of the research performed in Moyer and Howell [6] using LW and SW imagery; however the experiment where LW and SW source band imagery was presented to each eye of the observer in parallel could not be completed due to the effects of binocular rivalry [1] caused by the competing information presented in each eye independently. The remaining experiments performed using the LW and MW imagery were repeated in this work using LW and SW imagery. This paper is outlined as follows: a background section describes some common approaches to image fusion along with some common image quality metrics and their shortcomings regarding predicting human task