The SPACE (Sun, Precipitation, Atmosphere, Clouds, Earth) Thermal Signature Model has been developed by XonTech as a tool to be used in the accurate prediction of military thermal signatures. Currently this model has been optimized to address 8-12 micrometer signatures of armored ground targets in natural background settings. With somewhat lesser accuracy the current model design can address the 3-5 micrometer spectral region. With some model modifications, air and space targets could be addressed. The model is based entirely on first principles with respect to the thermal signature components induced by the natural environment. However, self-heating effects such as those caused by a tank engine or by friction require empirical input data which must be derived from pre-existing thermal measurements. The SPACE model has been programmed in compiled Microsoft BASIC to run on PC-compatible computers. Some generic target and background descriptions are part of the model ensemble. The development of additional descriptive data bases to cover specific target/background scenarios is possible using related utility software which has been developed for this purpose. The SPACE model is currently being used both by Government and industry to support model comparison studies, the prediction of target-to-background thermal contrast signatures, and the generation of synthetic infrared thermal imagery. It is the purpose of this paper to provide a brief tutorial on the modeling principles behind SPACE, a description of the SPACE software architecture and operation, and some example problems.
An analysis of clutter in IR background images was performed on image data videotaped in the southern USA and on a variety of IR image data from other sources. Images collected in the southern USA area, the Rome NY area and in central Europe were included in the study. The clutter in the images was computed as the rms noise in the pixel levels over the image. The images were subjected to spatial filtering procedures and the clutter calculations were repeated on the filtered images. Spectral densities were computed for the images in both filtered and unfiltered states to obtain information on the spatial frequency content of IR images of background scenes. The computed clutter values were entered into a data base and sets of data were extracted and plotted to determine the influence of parameters such as weather, scene type and measurement conditions on the clutter level of an image.
A large collection of radiometric signature data on various military targets was utilized to develop a simplified multiple linear regression technique for signature extrapolation. Environmental considerations were limited to air temperature, solar radiation, sky temperature, and wind speed. Targets were broken into about ten distinct classes of materials for predictive purposes. Results are presented for a number of the classes of materials found on typical targets. Data are compared with regression predicted values to evaluate the goodness of fit. Extrapolated signature values for a tank target are compared with predictions of two of the more comprehensive thermal models.
The clutter suppression capability of frame difference filtering using staring arrays is evaluated. With the assumption that the sensor platform is at low altitude, the efficacy of the filter on terrain clutter is determined analytically using the standard frequency domain approach. The dependences of the clutter suppression factor on such parameters as platform velocity, altitude, slant range, sensor field-of-view, and frame time are determined. The frame difference filter is also evaluated for a stationary sensor observing an explicitly time dependent clutter source.
In recent years, there has been vast interest in computer modeling and simulation of sensor imagery. Computer image simulation is an effective tool for the evaluation of sensor performance over a wide range of sensor parameters, scenes and weather conditions. In addition, simulated images of a given scene by various sensors under a given weather condition can be used for image fusion study.
Current aircraft have a requirement to operate at night and in adverse weather where optical imaging systems are inoperable. Imaging sensors operating at other wavelengths have the potential to provide vision through severe weather, but these systems need to be simulated before assuming the technological and financial risks involved in hardware development. Sensor and atmospheric models have been developed which simulate images at a variety of wavelengths. These models have been incorporated into a modified version of the IVEX Corporation Behold software which is used for the creation of three dimensional views of terrain data bases and includes fractal texturing and anti-aliasing. This new version, called Behold-ms, adds phenomenological models of material properties, such as surface roughness, emissivity, and temperature, and structured atmospheric weather models that consider path emission, backscatter, and specular/diffuse reflections of the sky. To date, images have been simulated in the visible (color), infrared (8-14pm), passive millimeter wave (35 GHz and 95 GHz), and active MMW (35 GHz and 95 GHz). These algorithms can be used for other windows over this spectral range. In order to accommodate the widely varying types of sensed energy while maintaining a practical amount of internal storage, a scheme for scaling each spectral band has been developed. Spatial resolution degradation due to diffraction, which is especially important at millimeter wavelengths, spatial sampling effects, and system noise models are also included. These sensor models and simulations have been used to examine adverse weather landing systems. Simulated images have also been used in image understanding research and spatial superresolution studies.
The Waterways Experiment Station (WES) has for several years been engaged in the development of numerical models for the simulation of terrain surface temperatures. Models have been developed to describe major components of the earth's surface such as grasslands, lands essentially bare of vegetation, such as deserts, and forested areas. These models combine first principles and semi-empirical techniques and use local surface meteorology and material thermal properties for input. Recently, WES has begun work on using these models to provide terrain radiance information needed for production of realistic computer generated infrared (IR) images. The approaches used in terrain temperature modeling are discussed, along with the structuring of the codes in a form amenable to scene generation requirements. Data needed by the models are outlined, and model predictions are compared to measured temperatures. Various approaches to account for the effects of atmospheric contaminants which may be applied to computer scene simulation are presented. Computer texture generation techniques based on thermal modeling as applied at WES are outlined. Application of models in an IR scene generation computer code developed by WES is discussed and example imagery presented. Possible extension of current work to other spectral regions and additional efforts for increased realism are also discussed along with possible approaches.
A broad class of procedural models, especially fractal models, for generating realistic synthetic terrain imagery have been adapted for use with Defense Mapping Agency (DMA) digital terrain data bases. By merging the actual terrain data with the procedural terrain models, very realistic and detailed scenes may be generated over a large area. DMA terrain data bases have typically limited resolution of about 10 meters or more. The procedural models allow effective data base resolution to be increased to 1 foot or greater. This increased resolution provides much higher scene content and real world fidelity for a variety of simulation tasks including flight simulation and sensor evaluation.
Realistic atmospheric and Forward Looking Infrared Radiometer (FLIR) degradations were digitally simulated. Inputs to the routine are environmental observables and the FLIR specifications. It was possible to achieve realism in the thermal domain within acceptable computer time and random access memory (RAM) requirements because a shift variant recursive convolution algorithm that well describes thermal properties was invented and because each picture element (pixel) has radiative temperature, a materials parameter and range and altitude information. The computer generation steps start with the image synthesis of an undegraded scene. Atmospheric and sensor degradation follow. The final result is a realistic representation of an image seen on the display of a specific FLIR.
Infrared imagery appears superficially similar to monochrome television imagery. An accurate simulation, however, must not only produce realistic images of self-luminous objects--rather than illuminated scenery--it must also replicate the visual anomalies of the imaging system. These anomalous effects arise from the imperfect nature of infrared imagers; they vary among systems, and they can become the dominant visual aspect of the displayed imagery. A strategy has been developed for producing high-fidelity simulated IR imagery in real time. The approach relies upon modeling techniques, which can create a database of infrared scenery derived from visual data, and upon a post-processor coupled to an existing image generator (IG), which will produce IR system-specific effects. A software emulation of the post-processor (PP) has been developed which permits evaluation of its projected performance, as well as facilitating tuning of system parameters in order to achieve realistic IR imagery. To complement these developments in high-fidelity IR simulation, a set of software tools has been developed to afford an efficient means of generating IR-specific characteristics for inclusion in the IG database. These tools blend the physics of the scene, atmosphere, and sensor with the requirements of the mission to be simulated and the 1G system to be used.
Digital scene modeling provides a powerful tool for the simulation of visual and infrared imagery because it allows control of all aspects of the scene. Specific targets can be defined and placed in arbitrary back-grounds, and specific atmospheric conditions and diurnal effects can be included and changed with relative ease, allowing a comprehensive study of factors critical to a real-world scenario. A major stumbling block to digital simulation has been the computation load required to process scenes containing a realistic representation of natural detail. The traditional approach to computer modeling of natural scenes represents natural detail explicitly, requiring a complex geometric data base which is costly to render. An alternative approach to modeling natural scenes is to use a small number of surfaces to define the major geometry of scene features and to use texturing to imply surface detail. Techniques developed at the Grumman Corporate Research Center use a few simple surfaces to define major scene features, such as hills, trees, and clouds, and a mathematical texturing function to define minor topographical detail by modulating surface shading and translucence. This technique pro-duces realistic visual images with much less computation than the traditional approach. Continuing research at Grumman in cooperation with the Keweenaw Research Institute and the U.S. Army Tank Automotive Command (TACOM) has extended the use of this technology to the simulation of infrared imagery by using the statistical characteristics of measured IR data to control textural shading on scene surfaces. This allows the simulation of visual and IR imagery using the same geometric data base and provides a cost-effective tool that can be integrated in a comprehensive target acquisition simulation system which will include missile dynamics, multi-sensor simulation, image processing, and pat-tern recognition.
Recognition of real world targets in complex backgrounds under variable environmental conditions and operating states is presenting severe chal-lenges to designers of automatic target recognizers. This is primarily due to the lack of identifiable statistical invariants in the target/background signature. OAO Corporation has available a sophisticated signature prediction capability that can be used in conjunction with a natural language description of the recognition context to determine both the features and the feature strengths that are appropriate for the specified recognition context. Our signature prediction capability can be used to design a context adaptive target recognizer based either upon classical pattern recognition principles or more modern but less mature learning networks such as found in emerging neurocomputers.
A FORTRAN program has been written to produce a high spectral and spatial resolution simulation of a scene viewed by a downlooking IR sensor. The scene generation method used allows the specification of a detailed scene at an acceptable cost in machine storage.
This paper presents four techniques that can be used to increase the effectiveness of Monte Carlo simulations of target-oriented optical (visible and infrared) imagery. Included are strategies to optimize looping structures and to simplify the simulation of algorithms that involve spatial moment calculations. Importance sampling for simulating low-probability events (e.g., miss or false alarm) is also discussed.
Procedures for the integration of Landsat Thematic Mapper digital data with DMA provided DTED and DFAD data sets are presented. Geometric correction issues in the merging of the data sets are addressed and examples are given. Presentation of the results of a data merge are displayed using a licensed perspective sceneration program GTVISIT.
An image algebra has been defined which is capable of expressing all finite image-to image gray level transformations. The purpose of this paper is twofold: (1) to prove the sufficiency of the algebra, and (2) to introduce the reader to the basic concepts of the algebra.
A gradient operator for detecting edges in imagery corrupted by noise is presented. Edge detectors that use masks to approximate derivative operations become imprecise in the presence of noise. Edge detectors which operate over small spatial scales (3 by 3) are ideal for edge localization and representing small detail; however, they respond equally well to high frequency noise and coarse textural information. Larger neighborhood operators (7 by 7) reduce the response to noise; however, edge localization and detail may be jeopardized. The Rule Based Composite Gradient Edge Operator (RBCGEO) interprets and integrates the responses of three operators each designed to extract edge information at a specific spatial resolution. The premise is that true edges have properties which will carry through several spatial scales while noise and other ambiguities will not. The RBCGEO uses a rule set to assign an edge magnitude, direction, and confidence of edge based on the composite response of the three operators. The algorithm and experimental results are presented.
In computer vision systems images are the primary inputs. In order to develop a robust vision system one needs to derive a correlation between the intensity values recorded in the images and the surface properties of various objects appearing in the scene. Image intensity values recorded in various spectral bands of a sensor are produced as a final result of a complex matter-energy interaction mechanism governing the illuminating source, the scene, the sensor and the intervening medium. In passive multispectral aerial images the illuminating source is the sun and the atmospheric column appears as the intervening medium between the sun and the scene for the incident beam and between the scene and the sensor for the reflected or emitted beam. In this paper an approach for compensating for the atmospheric transmitance and path radiance effects applicable for the reflective optical wavelengths is developed. It utilizes the LOWTRAN-6 package for calculating the required correction parameters. Applicability of this approach is demonstrated by correcting high resolution multispectral images.
This paper considers the design of an object cueing system as a rule-based expert system. The architecture is modular and the control strategy permits dynamic scheduling of tasks. In this approach, results of several algorithms and many object recognition heuristics are combined to achieve better performance levels. Importance of spatial knowledge representatiOn is also discussed.
Synthetic discriminant functions (SDFs) are used to construct digital filters for identifying and registering tank targets on 8-bit digitized forward, looking infrared (FLIR) imagery through correlation. The synthesis and testing of SDFs which control the shape of the correlation surface about the peak are discussed. This is effected by shifting the reference images in the + x and ± y directions and encoding these shifted images onto the filter. The advantages of shifting the reference images include sharper target correlation peaks, suppression of sidelobes on the correlation surface, correlation peaks that occur closer to the registration point of the target, and better overall target discrimination. The reference and test imagery consist of aspect-varying tank images, with a high degree of clutter resident on the test set imagery. Successive filter reference images are chosen on the basis of being quantitatively most dissimilar, in a correlation sense, than other images already resident on the filter. Filter construction techniques are discussed, and train and test set classification results are presented.
A deformation-tolerant method for classification/recognition of low resolution (FLIR regime) ship imagery is presented which employs statistical transformations and correctors based on concepts of fractal geometry. Fractal analyses, applied to specific classes of contours, present advantages of high recognition accuracy, position- and size-invariance and are suitable for microprocessor-based implementation due to low computational and storage requirements. The relationship between deformation-tolerance, superstructure geometry, and inherent transformation characteristics is presented in terms of image-plane distortions induced by out-of-plane ship rotation. Comparison algorithms using feature-space correctors derived from the fractal dimension are discussed in terms of classification and recognition success rates and computational load.
This paper describes enhancements made to an AGA ThermovisionR 780 IR Scanner to permit its use in determining transient thermal distributions induced in a resistive screen by a pulsed microwave source. Examples of data taken using both pulsed and continuous microwave sources are presented.
This paper deals with each aspect of converting thermal images into a displayable television format within the constraints of a hand held viewing system. We will consider the data collection technique used, including analog to digital (A/D) conversion criteria necessary to meet system performance goals. Also to be examined is the use of high speed static memories for data buffering and time base correction. The use of dynamic memories for frame storage and scan conversion will be investigated. Furthermore, a technique for frame averaging for noise reduction will be examined. System timing as it applies to television and scan conversion is presented, along with criteria for the video digital to analog conversion (DAC) for producing standard television output.
This paper presents the design and operational capabilities of a stand alone, or host computer driven, Single Instruction/Multiple Data (SIMD) systolic arrayprocessor system designed around the Geometric Arithmetic Parallel Processor (GAPP)TM chip. Use of SIMD techniques in the system allows the data throughput of this processor to be substantially greater than similar size systems with more conventional computer architectures.
With the emergence of new technologies for microcomputer-based products, relatively inexpensive, yet powerful image acquisition and analysis systems can be constructed for obtaining and processing infrared imagery. Recent product improvements in high capacity magnetic storage, optical disks, video frame grabbers, array processors, and image printers have made the desk-top computer a logical choice for sophisticated infrared image processing and field acquisition. The vast assortment of software products available for the personal computer allows the user to employ database management packages, window environments , and other tools that can enhance the performance or operation of the system. This paper presents a design of an infrared image acquistion and analysis system based around an IBM PC-AT that employs an assortment of state-of-the-art peripherals and software products.
Over the past several years, the authors have been involved in the development of the ERDAS image processing and GIS system, which has been ported to more than half a dozen different computers, ranging from an Z80 based microcomputer to large multiuser VAX/VMS installations. In this process differences in compilers, operating system capabilities and computer architecture has been overcome to provide a system which is uniform in appearance to the user as well as to the programmer. Identifying the functions needed to implement the system and developing these in a computer independent manner has made this process possible.