Date of Award
Doctor of Philosophy
Image fusion is the process of combining information from a set of source images to obtain a single image with more relevant information than any individual source image. The intent of image fusion is to produce a single image that renders a better description of the scene than any of the individual source images. Information within source images can be classified as either redundant or complementary. The relevant amounts of complementary and redundant information within the source images provide an effective metric for quantifying the benefits of image fusion. Two common reasons for using image fusion for a particular task are to increase task reliability or to increase capability. It seems natural to associate reliability with redundancy of information between source bands, whereas increased capability is associated with complementary information between source bands. The basic idea is that the more redundant the information between the source images being fused, the less likely an increase in task performance can be realized using the fused imagery. Intuitively, the benefits of image fusion with regards to task performance are maximized when the source images contain large amounts of complementary information. This research introduces a new performance measure based on mutual information which, under the assumption the fused imagery has been properly prepared for human perception, can be used as a predictor of human task performance using the complementary and redundant information in fused imagery. The ability of human observers to identify targets of interest using fused imagery is evaluated using human perception experiments. In the perception experiments, imagery of the same scenes containing targets of interest, captured in different spectral bands, is fused using various fusion algortihms and shown to human observers for identification. The results of the experiments show a correlation exists between the proposed measure and human visual identification task performance. The perception experiments serve to validate the performance prediction accuracy of the new performance measure. the development of the proposed metric introduces into the image fusion community a new image fusion evaluation measure that has the potential to fill many voids within the image fusion literature.
dissertation or thesis originally submitted to the local University of Memphis Electronic Theses & dissertation (ETD) Repository.
Howell, Christopher Leonard, "Visual Task Performance Assessment using Complementary and Redundant Information within Fused Imagery" (2010). Electronic Theses and Dissertations. 92.