The majority of images produced by optical and electro-optical equipment are destined to be viewed eventually by human observers. The human sense of sight itself exhibits a variety of imperfections of imagery (due to the dioptrics), internal noise (due to discrete image sampling at the retina) and a variety of image transformations in the neural networks preceding the perceptual levels of the cortex. In addition all these characteristics are progressive functions of retinal image position, whilst perceptual processes themselves are subject to both statistical fluctuations and influence from other factors. There are therefore many facets of imagery which interact with the human sense of sight, the importance of the individual interactions being largely dependent on the application to which the presented images are put. An approximate representation of preperceptual image formation and transformation by human vision will be presented. A unified approach to the modelling of visual performance will then be summarised which takes account of the various display/observer interface factors and attempts to allow for variability in perceptual processes. Finally it will be shown how apparently widely disparate approaches to visual performance modelling may be reconciled in terms of this unified approach.

Normal vision is a dynamic process, which depends on the constant motion of the retinal image that is produced by eye movements. Studies in which these natural motions have been eliminated, and in some cases replaced by artificial ones (under the control of the investigator), reveal important properties of the visual process. In this paper, we describe the effects of image stabilization on luminous and chromatic contrast sensitivity, explore the ways in which sensitivity varies with velocity, compare the effects of artificial image motion with those of natural eye movements, and report some surprising properties of the mechanism responsible for the residual detection of stabilized images.

Two hundred-fifty transparencies, displaying a new digital database consisting of 25 degraded versions (5 blur levels x 5 noise levels) of each of 10 digitized, first-generation positive transparencies, were used in two experiments involving 15 trained military photo-interpreters. Each image is 86 mm square and represents 4096² 8-bit pixels. In the "interpretation" experiment, each photo-interpreter (judge) spent approximately two days extracting Essential Elements of Information (EEI's) from one degraded version of each scene at a constant blur level (FWHM = 40, 84 or 322 μm). In the scaling experiment, each judge assigned a numerical value to each of the 250 images, according to its perceived position on a 10-point NATO-standardized scale (0 = useless through 9 = nearly perfect), to the nearest 0.1 unit. Eighty-eight of the 100 possible values were used by the judges, indicating that 62 categories are needed to scale these hardcopy images. The overall correlation between the scaling and interpretation results was 0.9. Though the main effect of blur was not significant (p = 0.146) in the interpretation experiment, that of noise was significant (p = 0.005), and all main factors (blur, noise, scene, order of battle) and most interactions were statistically significant in the scaling experiment.

Research has been conducted in Perkin-Elmer's laboratories over a several year period in an attempt to assess existing summary measures of image quality and to evolve new measures. A properly validated image quality measure would allow prediction of quality from an information content viewpoint that may be used in the synthesis, design, or evaluation of electro-optical sampled image systems. To further this goal, a set of psychovisual ranking experiments are presented and described in this paper. These experiments are intended to determine the influence of the sampling process, optical transfer function (OTF) shape, and OTF anisotropy on subjective image quality of sampled image systems. A description is given of the methods of experimentation employed, the procedure for providing a series of sensitometrically calibrated standards and test images for rank-ordering experiments, and the results of the tests that were performed. In addition, this paper summarizes an analysis of experimenter rankings to determine the degree of agreement (coefficient of concordance) among observers.

Information density and efficiency (i.e., the ratio of information density to data density) are used as criteria for assessing the quality of 2-D sampled and quantized imagery as a function of the statistical properties of random radiance fields, the spatial response (PSF or MTF) and sensitivity of imaging systems, and the sampling and quantization intervals. Computational results are intuitively satisfying: they are consistent with experimental and theoretical results obtained by earlier investigators concerned with the performance of TV cameras, and they provide useful guidelines for optimizing the design of line-scan and sensor-array imaging systems, especially if these systems use a digital communication link for transmitting data.

The operational performance of electro-optical imaging systems is currently predicted using models based on an approach introduced by Johnson in 1958. This model has been shown to be inadequate when used to predict performance over a wide range of conditions of signal-to-noise ratio and range to the target. This paper describes our efforts to develop and validate a model which is more effective than the Johnson model in predicting image quality. Implications in performance prediction modeling and electro-optic sensor design tradeoffs between sensitivity and resolution are also discussed.

The optical principles of the rigid endoscope are reviewed and a model for the assessment of image quality is developed.

Characteristics of radar scattering, the SAR imaging process, and display devices are considered with respect to their effects on the quality of the resulting image. A set of SAR image quality measures are specified which relate to application information requirements and associated measurement techniques are presented.

An approach, which utilizes Modulation Transfer Function (MTF), is used to develop tolerances for manufacture, assembly, and use of an optical system, and to predict the overall system MTF. This is accomplished by determining the MTF sensitivities for radius of curvature, thickness, and index of refraction for each optical surface. Each of these optical elements can then be toleranced for manufacturing and assembly based upon MTF sensitivity. In addition, the MTF effect due to tilt and decenter of each optical component can also be determined; this provides a bases for system alignment and focus requirements. The manufacturing effect for each optical component is converted to an MTF equivalent. The overall system MTF is then predicted by cascading the design MTF, focus MTF, alignment MTF, and manufacturing MTF.

Mirror surface ripple acts as a phase grating to diffract light out of the central maximum of the point spread function (PSF), reducing image quality. The effects of ripple on image quality are examined with the aid of computer simulations using rotationally symmetric wavefront error models, and through interferometric measurements of a mirror known to have significant surface ripple. Image quality is evaluated in terms related to the performance requirements of large orbital astronomical telescopes that must perform in both ultra-violet and visible light. Techniques for measuring and specifying ripple are discussed.

The Modulation Transfer Function Measurement System (MTFMS) is a microdensitometric system which incorporates solid-state scanning technology to produce data useful for image quality assessment of photographic images. The MTFMS consists of a two-channel microscope system mounted on a moveable-bridge light table. The background illumination is high enough to permit searching of the trans-illuminated specimen, and is augmented with a mechanical tracking light source which provides the illumination required for measurement. The microscope provides a binocular viewing port to the operator and simultaneously images the specimen onto a 100:1 aspect ratio self-scanned photodiode array. The instrument thus simulates a slit-scanning micro-densitometer, but does not incorporate a moving stage. The instrument is largely self-calibrating, keeping track of its own dark current and providing automatic calibration of each individual diode in gain and offset. A focus meter is provided and provision for parfocalization is incorporated. Integration times are selected automatically and data is corrected for these changes. The basic system includes a PDP-11/03 computer and associated equipment in a small rack which serves as a base for the operator terminal. This hard copy device can be used as an output printer as well as an input device and is useful for providing printer plots of output data. Software is provided for special purposes. Detailed testing of the instrument has been performed, demonstrating its densitometric performance and other elements of system quality. The results of these tests are reported.

The response of an optical system and its quadratic detector to an impulse in phase is expressed in terms of the amplitude PSF. This formulation is employed: Firstly, to indicate explicitly the conditions under which a phase point is rendered visible. Secondly, to extend the treatment to phase discontinuities to obtain an edge response in phase imagery.

Effects of tilt of a four-bar pattern on the minimum resolvable temperature difference (MRTD) of FLIR are discussed. When the bar pattern is tilted from the direction perpendicular to the detector scan direction, the displayed image of the bars has zigzag edges and the effective overscan ratio for the displayed image of the tilted bar pattern is reduced; both of these effects degrade the perceived signal-to-noise ratio. The quantitative treatments of these effects are presented. The perceived signal-to-noise ratio and the MRTD are de-rived as a function of the tilted angle of the bar pattern. To verify the theoretical results, MRTD measurements were performed with a serial scan FLIR which was developed in our laboratory. The overscan ratio of the FLIR was varied from 1 to 2 while the tilted angle of the bar pattern was changed up to 60°. Good agreement between the theoretical MRTD and the observed one was demonstrated.

The image quality of idealized dot formed line images is viewed from the perceptual attributes of width, width variation, and line sharpness. Expressions are developed which relate these attributes to the image variables of dot reflectance, dot diameter, and dot-to-dot spacing. Trade-offs are outlined between dot diameter/dot spacing, ρ, and dot diameter for these attributes. Regions are defined in ρ-diameter space where one or more of the attributes dominates the perception of these idealized lines.

"Acutance," an established concept, is an objective prediction of picture sharpness, which is subjective. "AMT acutance" is 100 + 66 log R, where R is the ratio of the area under the modulation transfer curve of a complete imaging system including a human eye to that under the eye curve alone, which is Gaussian with standard deviation 13 cycles per degree. AMT acutance 100 is excellent, 90 is good, 80 fair, and 70 just passable. We propose a new measure of acceptability so far as noise is concerned, on the same scale, to be called "granulance." It is based on the rms deviation, in 240-μm circular samples, from mean lightness in areas meant to be uniform. A tentative formula for granulance of a paper print is: minimum value of 100 - 40 log [650 x rmsL(film) x scene lightness x slope of the curve of paper lightness vs. scene lightness].

Although color is frequently used in computer graphics and presentation material such as slides and transparencies, it is still expensive to provide color prints. In many cases, such as quick proof prints, black & white would be far less expensive, and yet adequate, if colors were properly encoded as identifiable patterns. This paper describes methods for encoding colors as textures, preserving the ordering of lightness while generating recognizable patterns for colors.

A model is constructed for the noise-density characteristics of electrophotographic half-tone images. The model has components relating to the fluctuations: within the dot; within the background area; and in dot-size. These components are themselves related to the (toner) particulate nature of the image, enabling comparison to be made with the equivalent continuous tone case.

The various measurements necessary to test the model constructed in Part I of this paper are described. These measurements are used to calculate the model constituents of electro-photographic halftone noise, and hence to predict the overall noise. The predictions are then themselves compared with practical measurements of overall noise, and a close fit is found between theory and experiment. The fundamental role of toner particle size is confirmed.

The limitations of human vision provide the justification for design of systems for printing or display of halftone images that are perceived to be nearly identical to a continuous-tone original. Despite this fact, a review of existing halftone algorithms indicates that only limited use has been made of the extensive research on human vision. This review also emphasizes the common features of the many algorithms that have been proposed. It is suggested that iterative techniques may provide a means of incorporating a much more complete visual model within the halftoning algorithm.

Photographic image quality can be improved, both in theory and in practice, by the application of chemical techniques. These procedures result in the modification of the transmission or emission characteristics of the original negative. Transmission methods such as physical development increase the effective optical density of the original image by deposition of chemicals at or on the original silver grains. Emission methods such as autoradiography and fluorescence induce an increased optical density image on a new substrate. The information content of the intensified images is often greater than that of the original negative.

This paper is a preliminary report on investigations into the effects of data reduction on ISAR image quality and classifier performance. The case considered here is the effect of decimating the ISAR radar returns. Decimation of the type described induces image aliasing because of the signal processing involved in ISAR image formation. This familiar effect is used to model the degradation of object boundaries, which are used in classification.

This paper defines a system for enhancement of FLIR images, The algorithms presented are directed specifically toward the restoration of scanned infrared images and contrast enhancement for visual detection of targets. Image streaking is a problem common to AC coupled infrared scanners. Although it is not possible to uniquely restore an image degraded by this process, the algorithm attempts to reconstruct the most likely thermal scene. In essence, the pseudo-DC correction term is obtained from the line-to-line difference histogram. Since infrared imagery is limited in quality by photon shot-noise, the analysis emphasizes the algorithms performance in a highly noisy environment. For display purposes, the relatively large dynamic range signal at the sensor output should be simultaneously compressed and enhanced. A technique has been developed which accomplished dynamic range transformation while attempting to preserve several image properties crucial to visual recognition of targets. The algorithm exploits the range compression and automatic gain function of a global histogram modification, but does not ignore important spatial information.

Three aspects of image quality and their effects in an optical matched spatial filter correlator are described. These concern operation on: digitized imagery, data with low modulation and low space bandwidth product, and synthetic reference imagery. To address these practical problems, we employ: spatial filtering, edge enhancement, use of different apertures, photoreduced imagery, and digital preprocessed data. The first exuerimental data on optical matched spatial filter correlations with synthetic reference imagery and the first comparative data on digitally and optically processed multisensor image correlations are included. From these experiments, we find an optical weighted matched spatial filter correlator to be adequate for most multisensor data and that advanced digital preprocessing operators are necessary when presently available synthetic reference imagery is used.

For an image not dominated by regular features or structures, it is often desireable to quantify its "information content" in terms of an effective number of independent samples, N_i. This quantity is useful in several applications such as understanding the performance of certain image detection and registration schemes or calculating data redundancy. This paper offers a definition for N_i and applies it to three types of correlation functions postulated as spanning the range of "real world" image correlation: exponential, Whittle, and Gaussian correlation models. As an illustration of its use and importance, N_i is calculated for all three models at three distinct stages of a matched filter image detection system. Limited analysis of one type of aerial imagery supports the hypothesis that a Whittle model may be the canonical form of image correlation.

A major problem which has plagued image processing has been the lack of an effective image quality measure. It is well known that common measures which are mathematical and analytically tractable do not correlate with human subjective evaluation. This paper presents the results of a subjective evaluation on twelve versions of a black and white image (the SPIE GIRL) and the rank ordering obtained with three computational measures. It was found that a measure based on a model of the human visual system compared to the subjective evaluation with a correlation of .92.

The fundamental assessment criterion for TV-reconnaissance images is the perceptibility of targets, e.g. vehicles or buildings. An approach to reduce jam susceptibility of the video transmission between a sensor, e.g. at a remotely piloted vehicle (RPV), and a receiving station is bandwidth reduction by data compression, either spatial (e.g. differential PCM (DPCM)) or temporal (e.g. frame rate reduction). The influence of data compression on image quality can be investigated either by objective or subjective evaluation methods. An objective approach is the measurement of the observer performance variation (detection rate, identification rate) at various coding and noise conditions. We made these measurements with two-dimensional (2-D) DPCM-coded static and dynamic aerial scenes for varying bit rates and noise simulations. It is suggested that, under the assumption of a noiseless channel, bit rate can be reduced up to 1.5 - 2 bits per picture element (pixel) without considerable loss in observer performance. In the case of a channel with constant noise power density ('white noise') one must compromise between a better image quality at lower data compression and an increase in the anti-jam capability at higher data compression. In our investigations we found hereby the observer's performance maximum at a bit rate of 1.5 - 2 bits per pixel, whereby it is remarkable that a reduction below 2 bits per pixel resulted in considerable amount of error responses in the case of dynamic scenes. The results of the objective measurements are compared with experimental results of a subjective evaluation method ('pairwise image comparison'). It is shown that the results of the subjective interrogations correspond to the data of the objective measurements. We conclude that for the purpose of decreasing jam susceptibility DPCM-coding enables reduction of the transmission bit rate up to 3 - 4 Mbit/s, relating to a compression ratio of 4 : 1 to 5 : 1, whereby image quality remains acceptable referred to objective and subjective quality criteria.

Hydroxy apatite (Ca10 (PO4) 6 (OH)2) which has a hexagonal crystal pattern is the main inorganic constituant of human bone, and in particular of dental enamel (1). A great controversy has arisen amongst biological scientists concerning the types of defects that lead to cavities formation (2,3,4,5,6). The methods of investigation have not so far made use of the wide potential of image processing. In particular the question of variation of periods of the lattice pattern has not been separated from the distortions that are due to the system (distortions due to electron microscope or to the preparation).