This paper provides a brief overview of the AFATL (Air Force Armament Laboratory) Image Algebra with applications to multidata/multisensor image manipulations.

Advanced sensor provide image with very high spectral and spatial resolutions. Analysis of such imagery requires understanding of the complete imaging process and an ability to utilize scene specific a priori knowledge in a systematic fashion. In this paper we present a multiresolution analysis approach for detecting objects in high resolution images. The approach utilizes general knowledge of the spectral properties of the objects in their detection. The approach is implemented and results of experiments show the feasibility of this approach in detecting objects such as river, buildings, roads and vegetative covers in satellite and aerial images.

An initial pixel classification into both region class and interior or boundary designation is made based on an analysis of the distributions within the grey level cooccurrence matrices of an image. Local consistency of classification is then implemented by minimising the local entropy of region and boundary information. This is a robust way of simultaneously segmenting an image into texturally homogeneous areas with a thresholded edge map. Examples are shown using forward looking infrared (FLIR) images. The technique has been extended to cover any digital edge operator and an example is shown for the Spacek operator.

A hypothesis testing method for target screening is proposed. The method assumes that over a target sized window the target as well clutter probability density distributions are Gaussian. A double window generating algorithm scans the terrain, calculates the target PDF and correlates the PDFs of the target and clutter. Target detections are noted where the correlation between the target and clutter PDFs are small.

Adaptive clutter suppression and detection filters provide increased target and background discrimination capability to an IRST surveillance system by dynamically tracking and suppressing the background clutter environment. Simulation results with real IRST data are presented, both for single pixel and 3X3 target point spread functions, p.s.f.'s, for various sizes and applied in various order adaptive temporal/spatial/spectral filters, and suboptimal Markov, sparse covariance and LMS spatial filters. The filters are subsequently traded off in terms of performance and implementation complexity. Topics covered include a discussion of the normalized crosscorrelation technique employed for frame-frame registration prior to temporal and multispectral filtering and a method for counteracting the often encountered ill-conditioning of the clutter covariance matrix. Additionally, for the case of correlated background clutter, an expression is presented for the normalizer constant-false-alarm-rate, CFAR, loss as a function of desired false alarm rate and the expected mean and variance of the adaptive threshold estimator. In particular, for the case where clutter is distributed as a first order Markov process, the normalizer loss can be evaluated as a function of specified false alarm rate, background pixel-to-pixel azimuth and elevation correlation coefficients and normalizer shape and size. Techniques for reducing the normalizer CFAR loss are presented including decorrelating the data in the normalizer window, applying better detection filters and changing normalizer shape and size. These techniques are analyzed in terms of performance and implementation complexity.

This paper describes a technique for making an existing monostatic laser signature capability bistatic, using available monostatic data bases. A wavelength scaling technique gives the resulting code a potentially significant improvement in scattering theory, and an interesting observation of the Opposition effect" is discovered in the existing data.

Recognition of the river mouth is the key to detect the coast line and to discriminate it from the river side. It is an essential step in automatic determing the coast length. To realize this the spatial information) the correlation between the boundary pixels) the curvature and the curvature maximum must be taken into account. The algorithm proposed here has been used on some fractions of the seashore in North Shandong of China) and the results are satisfying.

An analytic model is developed for the geometric rectification of aircraft multispectral scanner data using information on the position and attitude of the aircraft. The model is implemented by constructing a set of cubic spline approximations to the aircraft position and attitude. The splines are computed using a generalized Newton's method to obtain a best fit to a given set of match points. Several tests were run using various numbers and sizes of segments, and several sets of checkpoints. The results are presented with analysis indicating a high degree of accuracy and reliability in the method.

One of the most challenging issues in today's world of geographic analysis and scene simulation is not the technology for analyzing or displaying the geographic data, but instead, the technology for deriving the databases that would support such functions in various regions of the world where detailed source material may not exist. The processing of spatial data has become commonplace due to the existence of low cost computer systems and the availability of spatial analysis software. Whereas, one was once only able to find true geographic analysis in research institutes and large architecture/engineering firms, now commercially available Geographic Information Systems (GIS) are being used by cities and local government, small business planning firms, state and regional government, and throughout a myriad of federal government entities including the military. College coursework in a number of disciplines now involves the modeling or analyses of spatial data on small computer systems.

A new criterion for resolvability of optical pictures is presented for the case that there were only photon noise in the pictures themselves in imagery. The super limit of two typical objects are calculated according to this criterion.

Traditionally, infrared-system-performance evaluation techniques have relied on detected signal-to-noise analysis for a single element of the scene. This approach includes the imaging-optical system; however, the geometrical optics effects have been deemed sufficient to describe the contributions of the optical system. The diffraction-based image-forming theory generally in use in the visible wavelength region is applied to the infrared system to assess its performance. The theory of incoherent image formation is applied to self-radiating bodies that have the wavelength-dependent radiance of a Planckian radiator. Within the traditional approximations in use in the infrared system design, a closed-form solution is obtained for the detected incidence. It is shown that the detected incidence may be expressed in terms of effective radiance, the only term included in the traditional infrared system design, and a correction term which is introduced by the diffraction effects.

The application of CCD imagers for airborne cameras require a long focal plane (>12000 pixels), high light gathering capability, high MTF, low noise and low non-uniformity between elements. Optical butting of focal planes using the "beam sharer" technique enable the manufacture of a line imager with a light gathering capability which is at least twice the light gathering capability of similar line imagers that utilize the classical "beam splitting" technique. Some problems regarding the evaluation of such line imagers are discussed: - Diffraction effects that influence the image quality at the butting zone. - CTF and MTF measurements techniques including problems of interpretation when square-wave targets are used.

The Infrared signature of a vehicle is dependent on its surroundings in two ways. First, the vehicle image perceived in an IR scene is directly related to the thermal contrast between the vehicle and its physical environment. Second, the vehicle surface temperature is determined, in part, by the energy exchange at its boundaries which includes the continuous radiative heat exchange with its surroundings. A vegetation background model has recently been developed to support the vehicle signature model, PRISM. The model is a first principles approach to predicting vegetation temperatures based on boundary conditions established by the surface environmental input. This paper will discuss the background model in terms of thermal physical behavior, model assumptions, and the model itself. In addition, model validation will be presented that shows simulation results that fall into the maximum/minimum bounds established by diurnal thermal imagery data gathered in support of this activity.

An important part of WES's remote sensing program is the development of electromagnetic energy signature maps that vary as environmental conditions change in an area. Such maps are useful for assessing electromagnetic sensor technology since the effectiveness of electro-optical sensors is dependent on weather and terrain background conditions. A model has been developed to predict the laser signatures of various types of terrain backgrounds under selected weather conditions. Experiments have been conducted to calibrate the laser scattering model. This model has been integrated with terrain elevation data and land use maps to produce color displays that indicate terrain areas exhibiting high or low laser scattering. Displays provide researchers with a tool for developing the background signature terrain maps needed to evaluate tactical sensor. technology and delineate specific features of interest.

The ability to generate polarized IR scenes is a significant advance in scene simulation fidelity. Ultimately, the polarization information in a simulated scene derives from the polarization properties of the source radiation and the optical properties of the materials that comprise the scene. Water and clouds are the background materials most likely to produce polarized radiances. This paper discusses the phenomenology of IR polarization in terms of Stokes parameters, a methodology for incorporating polarization into scene simulation software, and the generation and use of polarized material reflection data bases. Examples based on the GENESSIS software are given.

A thermodynamic model for predicting the behavior of selected internal thermal sources of an M2 Bradley Infantry Fighting Vehicle is described. The modeling methodology is expressed in terms of first principle heat transfer equations along with a brief history of TACOM's experience with thermal signature modeling techniques. The dynamic operation of the internal thermal sources is presented along with limited test data and an examination of their effect on the vehicle signature.

The synthetic imagery generation facility created in-house at the Center for Night Vision and Electro-Optics (CNVEO) was interfaced with the Sung Precipitation, Atmosphere, Clouds, Earth (SPACE) model by the Xon-Tech Corporation to create the Synthetic Imagery Generation Target Atmosphere Detector (SIGTAD) model where environmental and sensor properties are inputs and an image that is realistic in the thermal domain is the output. The turn around time is of the order of days. The method is demonstrated by a sequence of synthetic images of tanks in realistic surroundings over a diurnal cycle. Texture addition and surround heating algorithms enable the computer generations that are nearly indistinguishable from Forward Looking Infrared Radiometer (FLIR) imagery.

High fidelity IR scenes are needed to support the development of target detection and tracking techniques. It is necessary to have flexible and cost efficient testing methods to test the effectiveness of sensors and detection algorithms. While directly measured data can be obtained and used to vary sensitivity, dynamic range, spectral bands, and spatial and temporal resolution for evaluation of future sensor systems, synthetic infrared scenes can provide both radiometrically accurate information and exhibit realistic spatial variations. Two elements that contribute significantly to sensor performance are target radiance and background clutter. The IR scenes generated by a terrain, cloud, and target computer model provide for both of these elements. Many aspects of the phenomenology incorporated into each of these computer models have been validated. The scenes are raster images from a composite of target images inserted into a background radiance map. Multiple scenes can be linked together by using a common background and variably positioned target, thereby simulating motion of a target through a background. A series of scenes can relate to each other by performing parametric studies on various radiation sources in the scene. The scenes can be viewed with a predetermined timing sequence. This can simulate real time inputs to sensor systems.

The advantages of using a synthetic data base for developing Automatic Target Recognition (ATR) algorithms are well recognized. Primarily, a synthetic data base allows the economical generation of images that represent scenarios from which it is difficult or impossible to obtain real data. Furthermore, synthetic imagery allows direct control of image parameters, such as target signal-to-noise ratio, which aids ATR algorithm evaluation. This paper describes Texas Instruments' (TI's) Synthetic Image Generation for Automatic Target Recognizer Evaluation (SIGNATRE) system.

An ideal approach to computer generation of infrared imagery of synthetic scenes is described in which first principles prediction of thermal signatures is used at all key stages. This is followed by a discussion of current synthetic infrared imagery generation capabilities at Georgia Tech, with emphasis on existing modeling approaches and how they compare with the ideal approach set forth as a paradigm. Finally, example synthetic IR imagery is presented and discussed.

The GENESSIS computer code, developed at Photon Research Associates and available to the scientific community, generates spatially and radiometrica4y accurate two-dimensional Earth background images. Originally developed to assist in the design of infrared sensor systems, GENESSIS can also be used to "extend" measured sensor data to new, unmeasured experimental conditions. This paper summarizes the GENESSIS code, outlines the measured data extension process, and presents results of recent upgrades which enable GENESSIS to create imagery in the visible and near IR (0.4-2.0) spectral region.

A true perspective scene generation program has been developed along with supporting interactive geometry selection routines for a IBM/COMPAQ personal computer. The system is integrated with ERDAS image processing software so that an entire visual analysis project from database development to product generation may be performed on the PC system.

A background simulation code developed at Aerodyne Research, Inc., called AERIE is designed to reflect the major sources of clutter that are of concern to staring and scanning sensors of the type being considered for various airborne threat warning (both aircraft and missiles) sensors. The code is a first principles model that could be used to produce a consistent image of the terrain for various spectral bands, i.e., provide the proper scene correlation both spectrally and spatially. The code utilizes both topographic and cultural features to model terrain, typically from DMA data, with a statistical overlay of the critical underlying surface properties (reflectance, emittance, and thermal factors) to simulate the resulting texture in the scene. Strong solar scattering from water surfaces is included with allowance for wind driven surface roughness. Clouds can be superimposed on the scene using physical cloud models and an analytical representation of the reflectivity obtained from scattering off spherical particles. The scene generator is augmented by collateral codes that allow for the generation of images at finer resolution. These codes provide interpolation of the basic DMA databases using fractal procedures that preserve the high frequency power spectral density behavior of the original scene. Scenes are presented illustrating variations in altitude, radiance, resolution, material, thermal factors, and emissivities. The basic models utilized for simulation of the various scene components and various "engineering level" approximations are incorporated to reduce the computational complexity of the simulation.

Current interest in space based interceptors using IR sensors to home on and intercept ICBM's during early phases of flight make earth background scene generation in the opaque, infrared bands between 3.0 - 15.0 (μm) an important issue. This paper describes an approach for generating scenes in these opaque regions for arbitrary geometry and spatial resolution from measured data of fixed geometry and limited spatial resolution. Power spectral densities (PSD's) in the appropriate regions are constructed from various measured sources and used to synthesize images for a specific geometry by fractal continuation. This approach is particularly useful since it avoids the severe computational complexity and time penalty involved with a first principles approach. This is especially important for generating the large number of scenes required for ground testing high frame rate KEW sensors. An evaluation of the sensitivity of a typical acquisition algorithm to the accuracy of the background PSD's used to generate the synthetic scenes is also presented.

A capability for generating sky images containing both clouds and aircraft has been developed through concatenating a variety of models and matting theim to an Image Array Processor. A family of apparent images are generated, each at a specific wavelength, whose gray scale values are in terms of absolute radiometric units. These images can then be combined as weighted sums to represent the appropriate spectral distributions for a specific sensor of interest. For example, this simulator has been used to generate sky scenes as inputs to a color CCD camera emulation.

The presence of clouds affects most infrared (IR) military sensors. Foreground clouds degrade or occult target signatures and background clouds clutter a scene. Models used to assess or predict system performance must include important features of clouds: absorption, single and multiple scattering, thermal emission, partial transmission and spatial non-uniformity. Exact models which account for the details of cloud microphysics require large computation times and inputs which are difficult to obtain. Aerodyne has developed an approximate model which produces realistic cloud scenes in a reasonable amount of computer time. An average optical depth for the cloud is first calculated by use of LOWTRAN6 with specified aerosol optical properties. These properties are combined with a multiple scattering model which uses a two stream approximation. This model assumes that 1) the cloud layer is a parallel plane and infinite, 2) there is a constant single scattering albedo which may be wavelength dependent, 3) there is local thermodynamic equilibrium between particles and atmospheric gases, and 4) there is a uniform cloud temperature and emissivity. The result is an average cloud radiance spectrum along a specified line-of-sight. The line-of-sight may be up-looking or down-looking, and up to two cloud layers may be present. Spatial non-uniformities are incorporated by use of a cloud texture model based on a 1/f spectral shaping of spatial variations. The final scene including the effect of the atmospheric path from the cloud to the observer is in-band integrated and recorded as a grid of radiances with an associated depth map for use in a target/background interface model.

FLIR imaging systems produce imagery from infrared radiation. The optical, electronic and mechanical characteristics of the system determine many of the characteristics of the image. Simulations based only upon derivations of infrared irradiances from first principles generate ideal imagery that is not representative of FLIR imagery. A system has been developed for generating an approximation of ideal imagery, based upon simplified heat balance equations. This ideal imagery is then modified in various ways to simulate some aspect of the imaging system. The simulation executes in real time, and is suitable for both engineering and training environments.