Objective image quality measure derived from digital image power spectra

Abstract

An objective image quality measure based on the digital image power spectrum of normally acquired arbitrary scenes is developed. This image quality measure, which does not require imaging either designed targets or a constant scene, utilizes the previously known invariance property for the power spectra of arbitrary scenes. The measure incorporates a representation of the human visual system, a novel approach to account for directional differences in perspective (scale) for obliquely acquired scenes, and a filter developed to account for imaging system noise as specifically evidenced in the image power spectra. The primary application is to assess the quality of digital images relevant to the image task of detection, recognition, and identification of man-made objects from softcopy displayed versions of visible spectral region digital aerial images. Experimental verification is presented demonstrating very good correlation (r=0.9) of this objective quality measure with visual quality assessments.