Abstract
Different metrics has been used to quantify retinal image quality [1], including measures of optical quality on the pupil plane, metrics in the image plane based on the point-spread function shape, and metrics defined on the Fourier plane that account for the contrast loss at the different spatial frequencies. We propose a distinct kind of metrics taken from Information Theory [2] that allow us to compare the statistics of object and image and to know how much information about the input is present in the output. In particular, we used the ‘mutual information’, which measures the agreement between object and image, and the ‘Kullback-Leibler divergence’ (or relative entropy), which measures the distance between intensity histograms. Point-spread functions were calculated from wave aberration data of real eyes, and convolved with an object scene to obtain the retinal image. Studied objects included sinusoidal gratings of various frequencies and contrasts, Campbell-Robson chart, and natural scenes. Image and object were then compared by means of the information theory parameters for different aberrations. Mutual information and relative entropy performed well in accounting for the degradation that optical aberrations cause in the retinal image. Two applications were further explored: first, maximization of mutual information emerged as a robust procedure to estimate best image quality in the presence of aberrations; second, the effect of object windowing on visual performance was studied by applying these metrics with changing number of cycles in the stimulus. Information theory measures are a powerful statistical approach to study image quality and visual performance.
Supported by Grant FIS04-01018 (Ministerio de Educación y Ciencia, Spain).