Model of the second-order statistic of the radiance field of natural scenes, adapted to system conceiving

An optimal imaging system must fulfil its specifications without an expensive and undue quality. However the translation ofhigh-level human requirements into low-level system demands is not easy. As human specifications deal only with objects orscenes to be seen, the knowledge of these objects and its properties relevant to the information transfer through the imagingsystem is critical. As many imaging system quality criteria are based on the knowledge of second order statistical properties ofscenes or objects to be imaged, the goal of this paper is to show that it is possible to extract these properties from high-levelmission requirements.Keywords : Image Quality, Radiance Model, Power Spectral Density, Figures of Merit, Imaging System conceiving 1. INTRODUCTION The first natural step in the design of an imaging system is to define clearly the observation task it must fulfil. But there existssuch a variety in landscapes, that it is hardly thinkable to characterise an imaging system over a finite set of scenes. Thepurpose of this paper is to demonstrate that this problem can be overcome in the conceiving phase by the description of theobservation task in terms of the Power Spectral Density (PSD) of the scene. We rely on the working of the human visualsystem to show that this quantity is much more predictable than its real-world counterpart. It is shown that specific PSDshapes are associated with the images our visual system is macic to see. Indeed, the great diversity found in natural images canbe summarised in a few simple models of this quantity. The pertinence of these models is verified over various real imagesand the experience of several authors. The importance of the knowledge of the scene PSD is recalled through classical figuresof merit. At last, we show that these model parameters depend on simple scene characteristics, which in turn come naturallyfrom a well-defined observation task.