Abstract
The visual system can be highly influenced by changes to visual presentation. Thus, numerous techniques have been developed to augment imagery in an attempt to improve human perception. The current paper examines the potential impact of one such enhancement, multispectral image fusion, where imagery captured in varying spectral bands (e.g., visible, thermal, night vision) is algorithmically combined to produce an output to strengthen visual perception. We employ ideal observer analysis over a series of experimental conditions to (1) establish a framework for testing the impact of image fusion over the varying aspects surrounding its implementation (e.g., stimulus content, task) and (2) examine the effectiveness of fusion on human information processing efficiency in a basic application. We used a set of rotated Landolt C images captured with a number of individual sensor cameras and combined across seven traditional fusion algorithms (e.g., Laplacian pyramid, principal component analysis, averaging) in a 1-of-8 orientation task. We found that, contrary to the idea of fused imagery always producing a greater impact on perception, single-band imagery can be just as influential. Additionally, efficiency data were shown to fluctuate based on sensor combination instead of fusion algorithm, suggesting the need for examining multiple factors to determine the success of image fusion. Our use of ideal observer analysis, a popular technique from the vision sciences, provides not only a standard for testing fusion in direct relation to the visual system but also allows for comparable examination of fusion across its associated problem space of application.