COMMITTEE ON THE PEACEFUL USES
OF OUTER SPACE
REPORT ON EFFECTIVE RESOLUTION ELEMENT AND RELATED CONCEPTS
Note by the Secretariat
In the context of its consideration of classification of primary data relating to remote sensing of the earth by satellite, the Scientific and Technical Sub-Committee of the Committee on the Peaceful Uses of Outer Space noted that it was necessary to continue scientific research to provide a technical definition of spatial resolution and to determine what aspects of data, such as resolution spectral characteristics, polarization etc., might correspond to particular applications.
In this connexion the Sub-Committee had before it at its sixteenth session a study on the comparison of the resolution of photographic cameras and instantaneous field of view of linescan instruments (A/AC-105/204/Add.1) which is a continuation of a previous study of characteristics and capabilities of sensors for earth resources surveys (A/AC-105/204).
This document pointed out that the term "resolutions" cannot be used without a corresponding definition of the target. Thus, to say a system has a resolution of X metres on the ground is meaningless, unless the nature of the target and its intensities are also specified. Similarly, it has been noted that the instantaneous field of view (IFOV) alone does not adequately define the spatial response of a scanner system.
The conclusion of the above study, based upon theoretical as well as experimental results, was that for a scanner like the LANDSAT MSS, the "resolution" is about 2.4 x IFOV for low contrast targets and about 1.6 x IFOV for high contrast targets.
The above study also clearly demonstrated that no simple, meaningful figure of merit has been defined which would enable the direct comparison of different types of sensor systems.
Because of the many definitions of resolution, as well as the complexity of their relationship with modulation transfer function (MTF) and IFOV, the Scientific and Technical Sub-Committee requested the Secretariat to provide a report on the concept of "effective resolution element" (ERE) before its seventeenth session in 1980.
In order to prepare this report the Secretariat consulted with Mr. W. Murray Strome of Canada and Mr. A. P. Colvocoresses of the United States of America, who first saw the requirement for a new figure of merit for comparing remote sensing imaging sensors of different types and was responsible for initiating the development of a new concept which he called ERE. Consultations were also held with J. Troughton, J. N. de Villiers, P. N. Slater, L. L. Thompson,
F. Doyle, R. Welch and R. B. McEwan.
This report does not provide a finalized definition of a new figure of merit. Rather, it introduces some concepts which should be further developed. The Sub-Committee may wish to recommend that the International Society of Photogrammetryand the Society of Photo-Optical Instrumentation Engineers be requested to refine the concepts to provide precise definitions of the concepts introduced here.
1. The purpose of this study is to examine the possibility of deriving one or more alternative measures of spatial performance - for a remote sensing image system - which might be more meaningful in terms of realistic targets than the currently used terms "resolutions" and "instantaneous field of view (IFOV)" and which could be used to reliably compare the performance of different types of systems.
2. At this stage, it would be desirable to define two commonly used (and often misused) terms as applied to remote sensing imaging systems, namely: "instantaneous field of view (IFOV)"' and "picture element (pixel)".
3. "IFOV or instantaneous field of view is the angular subtense defined by the limiting detector aperture of a diffraction and aberration-free sensor systems." Commonly, the IFOV is expressed as the dimension(s) of the "foot-print" of the detector on the ground at a given instant.
4. "Pixel or picture element is the data sample in the output product to which a radiance value is assigned." It has dimensions which are not necessarily related to the sensor system parameters.
5. In attempting to develop a meaningful alternative figure of merit which can be used to compare various sensor systems, it became evident that two quite distinct concepts are required. The first is radiometric and should be based upon the minimum field size (or width) whose radiance can be measured sufficiently accurately. The second is related to spatial resolution and should be based on identifying features without requiring an accurate measure of the radiance involved.
6. Among the related geometrical performance criteria gaining widespread acceptance in the photo-optical instrumentation community are the point spread function (PSF), the optical transfer function (OTF) and the modulation transfer function (MTF). The MTF has already been described in paragraph 14 and figure 3 of document A/AC-105/204. The OTF is really a more complete measure of system performance in that it includes the phase relationship as well as the amplitudedegradation of the target signal as the frequency changes, i.e., the apparent change in position of the sinusoidal pattern as a function of its frequency. The PSF defines the apparent shape of a point target as it appears in the output image. The OTF is essentially the spatial fourier transform of the PSF. 1/
7. It is possible to estimate the OTF or MTF of spaceborne sensors using natural targets. However, the techniques are very complicated. .2/
8. While these system performance measurements provide a great deal of information about a system, there are practical difficulties in their use, especially in comparing sensor systems of different types, such as cameras and scanners. The prime difficulty arises from the fact that different types of systems have differently shaped (or OTF or PSF) curves. Thus, the performance of two dissimilar systems can only be adequately compared if the requirements curve, described in paragraph 14 and figure 3 of document A/AC-105/204, is known. This is not normally the case.
9. Some simple comparison tests have been proposed, such as the spatial frequency at which the output -modulation drops to X % (e. g. , 50 % or 5 %) However, there has been no agreement on an appropriate value for X. This is probably because one value for X will make cameras "look good", another will make scanners "look better", while yet a third choice will make television type sensors "appear best." Thus, the instrument designers tend to favour a choice which shows their own system in the most favourable light. Few, if any, comparison parameters have been chosen to provide a measure of performance which is meaningful to the civilian remote sensing user.
10. The term "resolution" suffers from similar practical difficulties, as it depends on the target contrast. Instrument makers prefer to use infinite contrast targets so that their systems appear to have the finest possible resolution. Such targets do not often occur naturally on the earth's surface, and are thus not meaningful to the user. But there appears to be no agreement on the contrast which should be used.
11. Remote sensing data are used in a wide variety of applications. For some of these. relative radiometric accuracy of rnultispectral systems is of paramount importance, while for others the ability to delineate spatial features is more important. Applications such as agriculture, forestry and water quality rely heavily on the spectral features of ground targets, while cartography, urban studies and engineering applications rely more heavily on spatial properties. There is a need for new, clearly defined performance parameters which will address both sets of requirements. On the other hand, sensor designers are concerned with both aspects of performance which may be expressed adequately for their purposes through the MTF, OTF or PSF concepts.
12. The basic radiometric accuracy requirement should permit accurate multispectral classification of data to be carried out over selected areas under given conditions. For example, an agriculturist needs to know the (relative) radiance of a given field to a certain accuracy in order to reliably identify various crops of interest. If the field is radiometrically homogeneous and sufficiently large, he will have no difficulty in obtaining many samples of sensor data with an accuracy limited only by the inherent radiometric errors introduced by the sensor and atmospheric characteristics. As the size of the field decreases, a point will be reached where not even one sample will be obtained with sufficient accuracy. The question of importance to him is "what is the minimum size field which will yield at least one sufficiently accurate radiance sample for classification purposes". It is important to note that this minimum size is not equivalent to any of the commonly quoted definitions of spatial resolution and is generally significantly larger.
13. The basic spatial resolution requirement involves the ability to locate boundaries and to resolve the finer details of features in a scene. For these tasks, the radiometric values of the features are of secondary importance. For example, cartographers need to be able to identify and accurately locate boundaries, roads and shorelines and to resolve features such as small lakes. Roads of sufficient width and/or with high contrast with respect to their surroundings will be relatively easy to identify, but as they are reduced in width and contrast, they fill become more difficult and eventually impossible to identify on the image. The question of importance to the cartographer is "what is the narrowest road of a given type that can be reliably detected, and how accurately can its position be determined". Neither the minimum road width, nor the accuracy of location are equivalent to spatial "resolution".
14. In an attempt to develop measures of the performance of imaging systems which would be more meaningful to the user than those which have been used by the instrument designer, two different concepts are evolving, both of which have been labeled "effective resolution element". One of these concepts is based upon the need to determine the minimum size of an object whose radiometric properties can be reliably determined. The second concept is that of defining a parameter which can be used to compare the abilities of different sensors in detecting, locating or distinguishing spatial features.
15. To differentiate between the two concepts, we will designate the first measure by the term "effective radiometric resolution element" or ERRE, and the second by the term "spatial detection effective resolution element".
16. All measures of the spatial performance of a sensor system which are in use today have been developed by the instrument designers in order to enable them to test their systems and ensure that their are functioning properly. Generally, the tests have been designed to make the performance of the system "look good". This is accomplished by using such measures as "resolution for infinite contrast targets in lines per unit distance" (not line pairs) in the descriptions of the systems.
17. While the measures used by instrument designers are accurate if stated in a complete form, they often convey the false impression to the user that it might be possible to measure or identify objects much smaller than would actually be the case.
18. For the user interested in determining the content of a field or area, a new measurement concept is required. For many user applications, the basic requirement is to provide sufficient radiometric accuracy to permit identification of a particular ground-cover over selected areas under given conditions.
19. The proposed concept of effective radiometric resolution element (ERRE) is based upon defining the minimum area of the earth for which a single radiance or reflectivity value can be assigned with at least a certain specified confidence that this value differs by no more than a certain specified amount from the actual radiance or reflectivity of the same area under the given viewing conditions. The following might serve as the framework for a definition of ERRE.
20. "Consider a field of area A and of arbitrary shape with uniform radiance centred within a larger area of the same shape with an area of B2A, also with uniform radiance. Both areas are Lambertian reflectors over the angular ranges of interest. The ERRE of the sensor system is defined by the dimensions of the smaller area A such that when the difference between the radiances of the inner and outer areas is at least of the full range of the instrument, the radiance of the smaller area can be measured to a relative accuracy of Y% with a confidence level exceeding Z%."
21. In the above definition., there are four undefined parameters: B, X, Y and Z. What are the practical and useful values for these parameters? For visible and near infra-red sensors 3/ a choice of B = 5, X = 30%, Y = 3% and Z = 95%-might be appropriate. For radar 4/it has been suggested that Z should be 80% and that Y should be expressed as 1 decibel, rather than as a percentage. (See also A/AC-105/251.)
22. For applications such as cartography, where it is important to precisely locate the correct position of features such as roads and boundaries, it has been difficult or impossible to evaluate the relative performance of different sensors through the standard specifications of IFOV or spatial resolution.
23. In an effort to overcome this difficulty, a second concept of ERE, labeled spatial effective resolution element (SERE), is evolving.
24. Although this concept is not fully developed, it provides an indication of the factors to be taken into account when attempting to cor-rare different systems. These factors include: surface area on the earth, or its equivalent on the image plane, the IFOV and the spread function resulting from optical diffraction and aberrations as well as electronic lag or smear.
25. Essentially, the concept of SERE is an attempt to determine the IFOV of the hypothetical, unsampled (ideal) scanning system which would yield performance equivalent to that of any given real sensor system. Whether such an equivalence exists is yet to be determined.
1/ See "The Development of Image Evaluation Methods", H. H. Hopkins, Proc. SPIE, vol. 46; Image Assessment and Specifications, 1974, "'Recommended Procedures for Calibrating Photogrammetric Cameras and for Related Optical Tests", P. D. Carman, International Society of Photogrammetry Commission I, 1960 and Section 6 Supplements, 1968 and 1976.
2/ See "Sensor Performance evaluation of the Skylab Multispectral Photographic Facility", F. J. Corbett, and "Measurement of the Earth Resources Technology Satellite (E-RTS-1) multi-spectral Scanner OTF from Operational Imagery", R. A. Schowengerdt, R. L. Antos and P. W. Slater, Proc. SPIE, vol. 46, Image Assessment and Specifications, 1974.
3/ An example based on effective radiometric resolution of a multispectral scanner is available for reference in the Outer Space Affairs Division Library.
4/ An example based on the concept of effective resolution element as applied to imaging synthetic aperture radar systems is available for reference in the Outer Space Affairs Division Library.