Design Considerations For Phase-only Correlation Filters

Abstract

The effect of variation of design parameters on the performance of phase-only and binary phase-only correlation filters is presented. Correlation performance is analyzed from the perspective of optimized extraction of a target from an input scene consisting of the target, a natural background, and additive noise. It was found empirically that the dominant factor affecting the filter performance was the target image background level and that the optimum choice depends on the relative values of the target average value and the background average value in the input scene. The behavior is explained analytically by describing the target object in terms of a target silhouette function that specifies the target region, and a target detail function describing the target variations within the silhouette. The optimum choice for the target image background is shown to be that which results in the addition (rather that subtraction) of the detail and silhouette autocorrelations.