Abstract
We introduce a new adaptive image fusion algorithm to address the decomposition level problem in the multiresolution fusion technique. The fusion scheme consists of two parallel processes which are linearly combined on a spatially adaptive basis: lowpass fusion to exploit the inherent target-background discrimination capability of infrared sensors and highpass fusion to capture texture and shape details leveraging the visible band imagery. The weighting coefficients for each process are determined by maximizing a no-reference fusion quality metric. We demonstrate the effectiveness of the proposed algorithm against a variety of practical infrared fusion tasks where it delivers tangible performance gains relative to several well-known state of the art fusion techniques.