In imaging systems in moving vehicles such as tanks, planes, ships, rpvs, etc., resolution is usually limited by motion blur. Successful restoration of blurred images depends primarily on the knowledge we have about the degradation process parameters, i.e., the point spread function (PSF) of the blur and the noise statistics. For many blur situations, such as those deriving from uniform or sinusoidal motion, the most important parameter for proper identification of the PSF is the blur extent parameter. This parameter is the smear size in the blurred image of a point object in the original image. A new method for identification of the blur extent in an image, blurred by motion of the camera during the exposure time, is presented in this paper. The blur extent identification is determined from the motion-blurred and noisy image itself. The identification method developed here is based on a concept that there exists a correlation between the pixels along the blur of the original unblurred pixels. In order to expose the hidden information about the blur extent, the blurred image is first transformed to a secondary image that is more informative about the blur extent, and then the blur extent is extracted from this image using correlation attributes in the image. This permits fast high resolution restoration when the blur extent can complete our knowledge about the blur properties.

This paper examines the applicability of computational vision models (CVM) to characterize thermal and visual imagery. A specific CVM model is described for the analysis of individual target characteristics and background clutter. A unique feature of the methodology is the spatial and temporal decomposition of the input image into various bandpass filters or channels. A description is given of the various model processes along with some representative examples of the subsequent analysis.

One of the primary inhibitory factors for resolution of automatic target recognition (ATR) performance problems has been the inability to quantitatively characterize low signal-to-noise (SNR) target detection and classification algorithms, especially those which are challenged by high spatial frequency backgrounds. The preceding work addressed obtaining classification statistics and geometric pattern referencing characteristics with the target mean intensity distribution commensurate with the background intensities. The current effort maintains a similar approach; however, the ratio of target-to-background intensity is significantly reduced. This is achieved by increasing the obscurant's ratio of differential scattering cross section-to- total cross section (albedo). The objective is to establish 50 percent of the edgels (target edge pixels) on the target at maximum sensor-to-target range in the presence of high spatial background frequencies, including obscurants. In addition precipitation rate and range, as well as variation in obscurant albedo, are assessed. Since scenario dynamics is sought, no attempt is made to resolve target edgels as a function of a single variable, for example precipitation. All variables are allowed to vary independently. The synthetic smoke generated for these plates incorporates the combat obscuration model for battlefield induced contaminants (COMBIC). The target and background imagery is taken in the LWIR by a Keewenaw Research Center (KRC) TMI FLIR. The final images are morphologically processed, segmented, high SNR scenes. The findings are that the target set need not be of a higher intensity than the surrounding imagery, as required in many matched filter operations; the target need only possess a higher intensity gradient than the background clutter and obscurants. Smoke and obscurant intensities may be significantly reduced, or even removed, by this type of morphological image processing.

This paper presents a simulation and comparison of two different infrared (IR) imaging systems in terms of their use in automotive collision avoidance and vision enhancement applications. The first half of this study concerns the simulations of a `cooled' shortwave focal plane array infrared imaging system, and an `uncooled' focal plane array infrared imaging system. This is done using the United States Army's Tank-Automotive Research Development and Engineering Center's (TARDEC) thermal image model -- (TTIM). Visual images of automobiles as seen through a forward looking infrared sensor are generated, by using TTIM, under a variety of viewing range and rain conditions. The second half of the study focuses on a comparison between the two simulated sensors. This comparison is undertaken from the standpoint of the ability of a human observer to detect potential (collision) targets, when looking through the two different sensors. A measure of the target's detectability is derived for each sensor by using the TARDEC's visual model (TVM). The authors found the uncooled pyroelectric FPA to give excellent imagery and, combined with the advantages of the 7.5 - 13.5 band in the atmosphere and the higher blackbody exitance in the 7.5 - 13.5 band, the 7.5 - 13.5 uncooled sensor is therefore the better choice for imaging through numerous atmospheric conditions compared to the 3.4 - 5.5 cooled sensor.

In searching a field of view for an object of interest, observers appear to alternate between two states: wandering (rapid saccades) and examining (focusing on an attractive region). This observation is made based on eye tracker measurements and is consistent with the model proposed by J. F. Nicoll, which describes search as a competition between points of interest for the observer's attention. In this paper search is represented as a random process -- a random walk in which the observers exist in one of two states until they quit: they are either searching or wandering around looking for a point of interest. When wandering, the observers skip rapidly from point to point. When examining, they move more slowly, because detection discrimination requires additional or different thought processes. An interesting consequence of the two state approach is that the random walk must have two time constants -- the time constant for fast (wandering) movements and a different time constant for slow (examining) movements. We describe a technique which can be used to separate raw eye tracker data collected in a search experiment into the wandering and examining states. Then we postulate the relationship of the probability of wandering (or examining) to the attractiveness of the image. We use a clutter metric to estimate the relative attractiveness of the target and the competing clutter. We find that the clutter metric predicts fairly well the time spent in the two states.

A program named `MRTTOL' has been devised to help in evaluating the performance of forward looking infrared (FLIR) systems in the presence of tolerances. It is essentially a shell program to execute the NVESD analysis program `FLIR92' multiple times with randomly varied input parameters, then calculate statistics on the minimum resolvable temperature difference (MRTD) predictions of FLIR92. The input to FLIR92 is a template-style data file in which the data must be in a certain sequence within each of a set of data blocks. This lends itself well to setting up a file of tolerances with a one-to-one correspondence with each of the datum inputs. The MRTTOL program expects just this correspondence between a `nominal' data file and a `tolerance' file giving the expected variations on each of the datum in the `nominal' file. The program interprets each tolerance as being the 2-sigma variation of the associated datum in the nominal file. It repeatedly sets up a perturbed data set by randomly varying each input value, runs FLIR92 with the perturbed data set, and accumulates the horizontal, vertical, and 2-D MRTD values in an output file. After running whatever number of perturbed data sets the user calls for, the program calculates the mean and standard deviation of the MRTD at each of 20 equally spaced spatial frequencies (increment set by the user). The output lists the expected values of MRTD for cumulative probabilities at several percentages from 50% to 3-sigma.

An unofficial modification to the U.S. Army's ACQUIRE v.0 range performance model is presented. This modification allows a two-parameter atmospheric transmission model to be invoked at the user's discretion that more accurately represents broad-band atmospheric transmission. Calibration of the two-parameter model to MODTRAN7 predictions for several standard atmospheres in different LWIR and MWIR wavebands is performed. Evaluation of differences in range prediction between Beer's Law and the two-parameter model is done.

This paper describes an analytic, end-to-end, IR seeker/sensor dynamic performance model developed to facilitate system level design trades and performance analyses for proposed IR missile seeker/sensor systems. The model has been extensively validated against actual simultaneous dual-band IR imagery (midwave 3 - 5 micrometers , and longwave 8 - 10 micrometers ), collected so as to emulate tactical airborne seeker engagement geometries against a variety of backgrounds. Over 20,000 SIR comparison measurements, with both injected and real targets, have been made. Agreement between predicted and measured SIR is typically within 2 - 3 dB, over a wide range of target brightness, background types, atmospheric conditions, and processing algorithm approaches.

This paper describes the validation of an end-to-end IR seeker/sensor dynamic performance model. The model was developed to perform system level trades and design analyses for IR missile seeker/sensor design, and its structural framework was patterned after the physical flow of information through a user specified electro-optical system and missile engagement scenario. The validation methodology was based upon comparisons of performance model predictions of signal-to-interference ratio (SIR) relative to the performance measurements of a baseline algorithm stream, which processed actual dual-band IR imagery, emulating an actual seeker and processor. This metric was chosen for the validation study, since this is the critical parameter determining missile seeker acquisition range and tracking fidelity. In the validation study, over 15,000 SIR comparisons were made, as a function of various critical system parameters, including registration accuracy and fixed pattern noise level. On average the performance model predictions of SIR matched those of the baseline algorithm stream within 3 decibels (dB).

Search and target acquisition (STA) is a complex task involving the ability of a human, using unaided eye or aided optics, to search an area of interest and ultimately discriminate a target from its background. As the backbone of many system performance and force-on-force models, the STA process must be tested and validated to some degree of accuracy. The Army STA modeling community has conducted a model improvement program and is in the process of validating the new models. Standard statistical techniques, which test for perfection (e.g., H0: a equals b), are inappropriate for validating these models which do not claim perfection. This paper proposes a nonparametric (rank based) method for validating models to a specified accuracy [e.g., H0: (b - .1) < a < (b + .1)]. The interaction between the power-of-the-test and sample size is explored, and an example using a thermal sensor target acquisition model is used to illustrate the technique.

Carrier transport in signal processing in the element (SPRITE) detectors is an important phenomenon because it determines properties such as the responsivity and the modulation transfer function (MTF). The previous literature has presented approximate solutions to the transport problem that neglect boundary effects, which have long been thought to play a major role in SPRITE behavior. In this paper we present a new solution to the problem through the use of modal analysis. This method intrinsically includes the three dimensional boundary conditions, and thus is more complete than the previous analysis. Through the analysis, certain dimensionless numbers arise which can be used to characterize SPRITE structure parameters and clarify how these parameters impact device performance. Further, we use this solution to derive an expression for the MTF. The effects of the boundary conditions and the device geometry on MTF are investigated using the new theory. Our results show that the quality of the passivated surfaces has a weak influence on the MTF; however, the device width affects the MTF performance more strongly when the passivation is poor. The effect of blocking at the contacts is investigated. It is found that while marginal improvements in MTF roll-off could be achieved by making the contacts highly ohmic, the maximum signal amplitude is obtained with partially blocking contacts.

The use of focal plane arrays introduces an aliasing problem which transfers higher spatial frequencies onto lower frequency regions and restricts the available spatial resolution or field of view. An anti-aliasing interpolation algorithm has been developed to improve this spatial resolution. The image is initially expanded to the required size with a usable lower frequency region which contains the aliased higher frequency components. A maximum likelihood criterion is then applied to estimate this higher frequency information based on a statistical image model, so that the aliasing effect can be reduced.

Performance characteristics and test procedures for focal plane arrays (FPA) are defined. These definitions are generally compatible with those commonly used for single-detector elements. The performance characteristics include on-chip electronics for signal processing and readout. Time delay and integration (TDI) arrays are considered as well. Optical components other than focusing optics for spot and line illumination are excluded. The test procedure for TDI arrays includes the scanner. Figures of merit for individual detector elements of the FPA and for the FPA as a whole are derived, e.g., an effective array element detectivity, an array enhancement factor, a resolution figure of merit based on MTF measurements, and a uniformity figure of merit. A correction scheme for nonuniform response of the detector elements is presented together with a figure of merit for the correctability.

The choice of waveband for thermal imaging depends not only on the characteristics of the target, background and atmosphere, but on imager parameters such as electron storage capacity per pixel, frame rate, optical transmission, cold-shield efficiency and non-uniformity. This paper outlines the ways in which the conventional trade-offs of sensitivity against imager and scene parameters are modified when electron storage capacity and non-uniformity are taken into account. In the presence of these limitations, the emphasis is frequently on optimizing image contrast rather than signal level or even photon-limited signal/noise ratio (SNR), resulting in more severe degradations due to equipment limitations such as transmission of hot windows or cold-shield inefficiency, and in reduced spectral bandwidths for optimum sensitivity. The dependence of sensitivity on cut-off wavelength is illustrated for a range of imager parameters, and the benefits of cooled or pseudo-cooled spectral filters are discussed. Optimum cut-off wavelength, particularly in the 3 - 5 micrometers (MWIR) band is found to depend strongly on system parameters. Comparisons are made between scanning and staring systems. In the 8 - 12 micrometers (LWIR) band scanning systems can compete well with staring arrays for equivalent frame rates, but improvements in multiplexer technology to give higher storage capacity or faster read-out will increasingly favor staring systems for all bands.

Traditionally single detector radiometric cameras have been employed to characterize target and background signature statistics during field tests. In the case of captive flight testing, several problems are encountered with this technique such as the limited availability and added cost of a stabilized radiometric camera payload and the undesirable scanning effects resulting from the moving platform. To overcome these problems during previous captive flight tests (CFTs) of imaging infrared (IIR) staring seekers, target and background temperature statistics have been extracted directly from the digitized seeker imagery and calibrated using blackbodies positioned in the field during the test. The question of how results of this method compare to those of the traditional method of using a single detector, radiometric system should be answered to evaluate the accuracy of the calibration achieved. During a CFT conducted at Redstone Arsenal, Huntsville, Alabama during July 1994, a pair of IIR, focal plane array (FPA) seekers were captive flight tested and signature statistics were extracted from the imagery. As a point of comparison, imagery was coincidentally collected using Agema 780 IR radiometers during a single mission. Calibration was performed by using large surface blackbodies placed adjacent to the target so as to appear in the field of view of the sensors as the target was being imaged. This paper presents the results of the comparison between the seeker and radiometer calibrations in both the mid-wave IR (MWIR) and long wave IR (LWIR) spectral bands. The results have application and relevance to IIR model validation through field testing.

Imaging IR devices, particularly two-dimensional staring FPAs, are capable of providing large amounts of data applicable to target detection and tracking. This information is received as dynamic imagery of the subject scene, with commercial sensors currently delivering digital data at rates above 35 megabits per second. Currently, due to these high data rates, system performance is limited by the processor throughput and performance. Efficient methods are needed to suppress the deleterious effects of clutter and noise in order to achieve operability against low-signature targets in high-clutter conditions while maintaining real-time operation. To accommodate this, fundamental Wiener filter concepts were applied to design several digital processing concepts. These concepts used noise models that were based on recent progress in representing background clutter. Although not implemented, the general technique is applicable to additional disturbances such as fixed-pattern noise, assuming the noise power spectrum is known. The IR background was modeled using a three-parameter Gauss-Markov power spectral density. Four digital processors were designed using finite impulse response (FIR) kernels of varying size, assuming a fixed-size circular target in each case. A simple band-pass kernel tuned to the target was also considered as a baseline non-Wiener filter. These filters were compared based on their realized transfer functions compared to the ideal Wiener transfer function. The detection processing was accomplished by convolving single-frame imagery with the FIR kernels. Using the filtered imagery, varying thresholds were applied to derive the receive operating curve relating detection and false-alarm probabilities. The processor details and design parameters derived from the noise and clutter parameters are described. Empirical and analytical assessments of filter performance were obtained in clutter backgrounds. Applicability to real-world IR backgrounds is demonstrated through evaluation against real scenes. The analytical model supports assessments of processor performance against extended targets, including quantifying the performance penalties due to imposition of realistic processing constraints.

Current performance measures for thermal imagers relate only indirectly to human recognition. However, as the probability of recognizing an image may be used as a measure of image quality, the relationship described in this work could contribute to the formation of a new objective performance measure for thermal imaging systems. The human recognition probabilities of several degraded shapes have been evaluated using computer generated images displayed on a computer monitor. The shapes investigated were blurred by convolving them with a range of two dimensional Gaussian functions. In subsequent trials, these blurred images were further degraded by sampling (in order to simulate the effect of detector arrays in an imager). Images of degraded shapes were presented to observers in a random order and with a random degradation. After each presentation the observer decided which was the most likely shape to represent the image displayed on the screen. A correlation has been found between the probability of recognition of a particular degraded shape and the relative contrast between the image of that shape and a similarly degraded circle of the same area. This is true whether the degradation is due to blurring alone or to blurring followed by sampling.

Correction of photoresponse nonuniformity in infrared staring sensor arrays is investigated. A general nonuniformity correction procedure is proposed. The procedure is based on multiple irradiation sources and on least square fit approximations to the individual pixel response characteristics. Nonlinearities of the signal response are taken into account. A correctability figure of merit is defined which may be used to estimate the residual fixed-pattern noise after correction. The correction procedure and the correctability are applied to a real data set measured by a 64 X 64 element infrared focal plane array (FPA). It is shown that an offset correction is insufficient for this data set and that a linear correction reduces the fixed- pattern noise contribution to the magnitude of the temporal noise background. The residual uncorrected fixed pattern noise can be related to pixels showing large temporal noise.

IRTool is an IRST X Windows analysis tool, which is being developed by Arete Associates and NSWC/WO under the sponsorship of the Office of Naval Research in support of the Infrared Analysis Modeling and Measurements Program (IRAMMP). The tool consists of an integrated set of physics based modules to support IRST multispectral and space-time analyses. The primary modules are for (1) modeling atmospheric effects, (2) simulating ocean and cloud scenes without and with sensor effects, (3) modeling and injecting target signatures into real and simulated data, and (4) analytic calculation of the expected signal-to-noise ratio (ESNR) for an airborne target on a specified trajectory. Additional modules support data processing and analysis for clutter characterization and model validation. These modules have undergone extensive verification and comparison with data. IRTool has an interactive X Windows driver, which launches stand alone modules to run in the UNIX background. The user can interactively display and plot module outputs using IDL programs written for IRTool. IRTool is available from the IRAMMP program manger (Douglas Crowder).

Absolute radiometric calibration, measurement and error analysis techniques are described for IR imaging radiometers as they apply to field measurements. A serial scanning type system previously used for relative (contrast) measurements is presented as an example in illustrating the techniques. Error sources that are unique to the absolute measurement problem are highlighted. Two field measurement examples are presented to illustrate how the individual error terms can vary and affect the overall measurement accuracy.

There is a growing demand for a larger field coverage for scanning in IR instrumentation. We have explored the possibility of using the microlens arrays (MLAs) for this purpose, and have developed a generic IR imaging scanner. The application of MLAs in the wide-field IR imagers is presented, along with their advantages and limitations. A series of systems using MLAs and diffractive elements have been designed and analyzed for a diffraction-limited performance with 45 degree(s) and 60 degree(s) field of view for a f/1.4 IR imager operating in 3 - 5 micrometers wavelength range for 1-D and 2-D scanning applications. The optical design considerations, fabrication issues, and thermal effects are also discussed for these types of scanners.

Automated test methods have been developed and implemented which provide a high degree of correlation with average manual test results using human observers. The results of this effort are given in this paper using the presently implemented MRT models. New data using the FLIR92 3-D noise model are also presented which offer a more detailed description of sensor noise over previous implementations. This improves the accuracy of automated MRT and will facilitate testing of scanning time delay and integration (TDI) and staring array thermal imaging sensors in the future.

A prototype automated forward looking infrared (FLIR) minimum resolvable temperature difference (MRTD) evaluation software system was developed and tested. After data capture and preliminary image processing of FLIR 4-bar target imagery, the boundary contour system (BCS) model of the human early vision system was coupled with a custom feature extractor to produce a set of features characteristic of those employed by humans during detection tasks. These feature sets, along with known target visibility, were used to train a fuzzy adaptive resonance theory MAP (ARTMAP) decision algorithm to emulate human observer performance in determining MRTD as a function of target to background contrast and target spatial frequency. During prototype system evaluation, the system was trained on 180 pairs of input imagery and human observer response data (resolvable/not-resolvable), and then tested against another 60 input images without the human judgments. The system predictions of human response to the test images were than compared to actual human response decisions for the images. Prototype success rates in the range of 96% to 100% were achieved in correctly predicting human response MRTD decisions in a low fidelity situation.

The 3-dimensional noise model developed by D'Agostino and Webb is rapidly becoming the standard metric used to characterize the temporal and spatial noise present in modern day FLIR imagery. This model separates the noise components into seven directionally dependent noise terms. These terms are isolated into spatial and temporal components dependent on their directional characteristics. A value in units of temperature is determined for each component. This approach gives substantial insight into the workings of the sensor and its nonuniformity correction capabilities. Without knowledge of the spectrum of each of these components though, the impact of each on the minimum resolvable temperature difference (MRTD) results may be misinterpreted or misconstrued. Knowledge of the spectrum of the 3-D noise is also imperative when the 3-D noise model is used as the basis for noise generation in sensor simulation algorithms. This paper presents a method of obtaining 3-D noise spectrum for each of the seven noise components by first isolating each noise term. Once the individual noise component frames have been isolated, the spectrum of that component is determined. The determination of the noise spectra of each of the components is then discussed. Sample 3-D noise spectrum data are presented.

Accurate measurement of the modulation transfer function (MTF) for imaging systems can be obtained by viewing a known target that is broad-band in the Fourier domain. A slit target is frequently used; however, undersampled staring systems cannot properly reproduce this type of target since frequencies above Nyquist are folded into those below Nyquist, resulting in the well known aliasing effect. Proper accounting of aliasing is needed to determine the true system response to frequencies below Nyquist. To alleviate the aliasing problem, existing methods use various techniques that effectively increase the sampling rate across the slit. These methods usually require some type of positioning device in order to accurately control the relation of the location of the slit with respect to the sensor array. This paper provides an alterative method in which a marginally resolved slit is canted with respect to the sensor array. MTF values can be determined at any slit cant angle greatly simplifying the laboratory procedures and requirements. The two-dimensional (2-D) discrete fast Fourier transform (FFT) of the canted-slit image enables separation of the aliased and non-aliased signal components. This allows the accurate determination of MTF for frequencies both above and below system Nyquist. Temporal averaging and reference image substraction are used to reduce the effects of system noise on the measurements. MTF values determined using this method are compared to calculated, idealized predictions for three undersampled medium- wave infrared (IR) systems. This comparison provides the validation of this procedure and demonstrates its general utility. The general improvement offered in measuring MTF for aliased systems is presented in terms of limiting accuracy.

In continuation of the authors previous work, this paper presents a complete two-dimensional model for a serial scan FLIR sensor. This model, developed using a commercially available scientific spreadsheet (i.e., MathCAD ver. 4.0), and taking into account of all the basic parameters of the sensor as well as the 4-bar target, can predict how any 4-bar target would appear on a display.

The detector noise is the key limit to any imaging system. The unique design of signal processing in the element (SPRITE) detectors provides measurable signal to noise improvement over ordinary detectors. The analysis and description of the noise in this type of detector is complicated by the fact that transport phenomena filter the noise spectra prior to readout. Previous analyses of the noise behavior used an approximate solution to the charge transport problem to produce expressions for the noise PSD, and yield predictions of a flat PSD at low frequencies. Measurements of real devices show a large 1/f behavior at low frequencies, but this has always been attributed to contact effects. In this paper, we use the results of the eigenmode solution to the transport problem to derive the PSD of the noise. We show that this analysis produces noise PSDs that have a 1/f dependence caused simply by the operation of the transport processes. The result is then coordinated with the similarly obtained signal response MTF to produce the frequency dependent signal to noise ratio. These are computed over several ranges of detector parameters with the intent of revealing their optimum values.

Refurbishment of the Arnold Engineering Development Center (AEDC) 7V high-vacuum cryogenic chamber has recently been completed. This chamber now provides a high-fidelity ground calibration and test facility for infrared surveillance sensors and infrared seekers. Although the chamber optical system was originally designed for diffraction-limited performance in the MWIR to LWIR wavebands, its actual performance allows near diffraction-limited testing into the SWIR over a 20-mrad-square field of view. With a test pupil diameter of 50 cm, 20 K temperature capability, multiple blackbody targets, excellent vibration isolation, exceptional optical performance, and supporting analysis capability, the AEDC 7V is a premier space environmental chamber capable of meeting sensor/seeker ground test requirements well into the next century. Optical collimation of the blackbody sources is provided by a long focal length, unobscured cassegrain. Installation, alignment, and performance testing of the all-metal cassegrain in the chamber were achieved early in 1994. This paper reviews the computational procedures required: (1) to define the final ground test configuration and system alignment after installation of the cassegrain; and (2) to create a computational model that reproduces the interferometrically measured wavefront performance of the installed collimator.

An advanced sensor test capability is now operational at the Air Force Arnold Engineering Development Center (AEDC) for calibration and performance characterization of infrared sensors. This facility, known as the 7V, is part of a broad range of test capabilities under development at AEDC to provide complete ground test support to the sensor community for large-aperture surveillance sensors and kinetic kill interceptors. The 7V is a state-of-the-art cryo/vacuum facility providing calibration and mission simulation against space backgrounds. Key features of the facility include high-fidelity scene simulation with precision track accuracy and in-situ target monitoring, diffraction limited optical system, NIST traceable broadband and spectral radiometric calibration, outstanding jitter control, environmental systems for 20 K, high-vacuum, low-background simulation, and an advanced data acquisition system.

This paper examines the specification requirements for the development of standard module scanning components to be used in conjunction with SADA I and SADA II sensor arrays. System-level design considerations are presented to identify a selection of components that is consistent with optimum use of the SADA technology. A limited-rotation electromagnetic actuator, used in conjunction with an angular position sensor and a digital controller, is shown to have the necessary performance and flexibility to perform the frame scan function for a wide range of airborne systems. System level requirements and specifications for an optional interlace scan system are also provided.

This paper describes a method for parametrically characterizing IR clutter for missile seeker applications in terms of a Butterworth model of the power spectral density (PSD). Traditionally, models of the PSD have been characterized by a Gauss-Markov correlation model, or a similar variant. These tend to estimate the overall spectral shape and integrated spectral power (rms) level very well. However, they tend to be dominated by large spatial features, i.e., low wavenumbers, and consequently, tend to underestimate the power present at high wavenumbers. While this is often not a factor in many remote sensing applications, it is a critical issue for missile seekers, as targets are often sub pixel in extent, and hence compete against clutter at high spatial frequencies. The clutter parameterization algorithm performs a fit to a one-dimensional profile taken from a radially average slice through a two-dimensional PSD computed from geometrically and radiometrically corrected airborne dual-band IR imagery. The parameterization is iteratively constrained to optimize the fit at high wavenumbers, and a two-dimensional isotropic model is computed. A description of the parameterization algorithm, as well as a synopsis of model fits to various clutter types are presented. Additionally, a table of recommended global PSD parameters for clutter characterization (by waveband) is presented herein.

Blackbody thermal reference sources are critical test facility components used for the evaluation of thermal imager systems. Extended source blackbodies are used at the Night Vision and Electronic Sensors Directorate (NVESD) for testing and evaluation of new generations of thermal imagers. Wider field of view and more sensitive systems especially require knowledge of the accuracy, stability, and uniformity of the thermal reference sources used for their evaluation. This paper demonstrates measurement techniques used to evaluate extended source blackbodies and their controllers using a sensitive, stable, and uniform thermal imager, the Amber Radiance 1.

The line spread function (LSF) is a classical figure of merit used to derive the modulation transfer function (MTF) of an imaging system. However this test cannot be directly applied to cameras based on an infrared focal plane array (IRFPA), as these sampled systems exhibit aliasing effects. ONERA has proposed a new technique called generalized line spread function allowing MTF assessment for sampled imaging systems. A testing of spatial frequency filtering becomes possible until at least twice the Nyquist frequency. The purpose of this paper is to show that this method is well-adapted to IRFPA cameras. In particular, the effects of bad a priori estimation of the characteristics of the camera are discussed. Secondly, we have carried on a laboratory experiment with a widespread infrared imaging system. The purpose is to demonstrate experimentally the advantages of this new method, not only in terms of performances but also in terms of simplicity of implementation and robustness. The infrared imaging system tested is a 512 by 512 Si:Pt IR5120C Mitsubishi camera equipped with a F/1.2 Angenieux lens.

For years there has been a dichotomy within the test community about imaging sensor resolution figures of merit; one camp favoring the modulation transfer function (MTF) while the other prefers the minimum resolvable contrast (MRC). This presentation will attempts to bridge the gap between these two test approaches and describe how rigorous mathematical techniques might be applied to test data reduction in order to streamline the test process. We can superimpose the MTF and normalized noise curves into one 3 dimensional graph. Through Fourier and statistical analysis, a streamlined test approach and objective analysis is now possible. The potential for repeatability and cost savings can be realized through a two phased test procedure which anticipates these powerful analysis tools.

The Army Night Vision and Electronic Sensors Directorate (NVESD) has recently performed a human perception experiment in which eye tracker measurements were made on trained military observers searching for targets in infrared images. This data offered an important opportunity to evaluate a new technique for search modeling. Following the approach taken by Jeff Nicoll, this model treats search as a random walk in which the observers are in one of two states until they quit: they are either searching, or they are wandering around looking for a point of interest. When wandering they skip rapidly from point to point. When examining they move more slowly, reflecting the fact that target discrimination requires additional thought processes. In this paper we simulate the random walk, using a clutter metric to assign relative attractiveness to points of interest within the image which are competing for the observer's attention. The NVESD data indicates that a number of standard clutter metrics are good estimators of the apportionment of observer's time between wandering and examining. Conversely, the apportionment of observer time spent wandering and examining could be used to reverse engineer the ideal clutter metric which would most perfectly describe the behavior of the group of observers. It may be possible to use this technique to design the optimal clutter metric to predict performance of visual search.

The implementation of an anticollision system on a car is presented. The vehicle is provided with two LIDAR systems integrated near the headlights, and a RADAR included in the calender, both used for active detection. Twelve infrared beacons have also been set up on the front part for telecommunication exchanges with surrounding vehicles or with infrastructure. An uncooled copolymer 32 X 32 FPA for passive detection has in addition been investigated for a further integration and is presented in this paper. This project is a part of the European PROMETHEUS-PROCHIP program which intends to make safer road traffic.