The current interest in applying hybrid techniques to data extraction from film is discussed and reviewed. The three basic operations of power spectral analysis, spatial filtering, and correlation are described. Particular emphasis is placed on how these operations can be carried out with an appropriate combination of optical with either digital (binary) or electronic (analogue) subsystems. Some examples will illustrate the procedure.
The real-time functional features of optically addressed electro-optical spatial modulators such as the Pockels Readout Optical Modulator (PROM) are described and applied to hybrid optical/digital processing of photographic information. The current and predicted performance of developed hardware and R&D devices for optical image sampling, incoherent-to-coherent conversion, stored image manipulation and computer-controlled Fourier plane filtering are presented. The utility in image exploitation of adjusting the dynamic range of a stored image under interactive computer control to effect background suppression, optical density level contouring, image inversion, spatial filtering and edge enhancement is illustrated graphically.
In this paper we suggest objective methodology for evaluating tradeoffs between optical/digital components in a hybrid processor. This methodology is based upon signal detection theory as a means of evaluating component performance in specific tasks. Two signal detec-tion tasks are hypothesized and discussed, and we demonstrate that an assumed advantage of a digital system, precision or low-noise, is less important when realistic data sources are considered.
The marriage of optical and digital processing techniques into hybrid optical/digital systems is a powerful and practical system architecture. It allows one to retain the high-speed and parallel processing features of an optical processor while achieving the control, flexibility, and data analysis advantages of digital technology. In this paper, we describe our new microprocessor-based optical/digital interface, its performance specifications, operations achievable, and present several examples of this new system.
A summary is presented of current work on applying the optical power spectrum to measurements of photographic images. Conclusions are that the technique is useful for characterization information content of images and is predictably sensitive to focal errors and contrast degradation.
Although the optical power spectrum (OPS) is widely used in pattern recognition and image assessment application, relatively little literature exists on the practical design considerations for OPS systems. This paper addresses some of these relevant issues. We illustrate the three standard configurations for OPS systems but limit our detailed attention to only one. We establish a limit on the space bandwidth product for neglecting the curvature of the transform surface, devise formulae for the laser power requirements, and consider the effects of film substrate thickness and scan center stability on system performance. We conclude with brief remarks on the effects of spectral and temporal coherence.
An overview is presented of major error sources inherent in the diffraction pattern sampling technique as used for power spectrum measurment. Topics include sampling errors in the measurement of diffraction pattern intensities, the normalization of the data and the interaction of the system limiting aperture with the resulting data. Phase effects and other sources of film noise are also treated. The purpose of this discussion is to acquaint workers in this area with the potential for error and to indicate the need for careful interpretation of the data.
The application of communication theory and linear systems concepts to the analysis of photographic images and imaging systems is well established. These techniques include the use of the Wiener spectrum to characterize photographic granularity and to determine the transfer functions for photographic imaging and printing; the standard methods of measurement that have been developed use essentially incoherent optical techniques. Recently, however, coherent optical processors have been applied to image analysis. A series of experiments is described which allows comparison of the spectra and the transfer functions generated by the incoherent and the coherent techniques.
Optical Power Spectrum Analysis Diffraction Pattern Sampling has matured during the last decade into a technique that now is used routinely by scientists and engineers in a wide range of disciplines. In this paper we discuss recent developments in solid state optical power spectrum analysis systems, nonparametric pattern recognition algorithms and applications of the hardware and software technology. Today's modern diffraction pattern sampler utilizes a solid state photodetector array, wide-band high gain monolithic amplifiers and advanced digital logic circuitry. It is readily interfaced to minicomputers or microcomputers that can control the required Materials Handling both for laboratory and industrial applications. As many as 64 sample point: in the diffraction plane may be easily acquired 500 times per second with this type of system. Thus, gathering large data bases for valid statistical pattern recognition experiments or implementing quality control systems in an industrial environment is relatively easy. Nonparametric feature analysis procedures for sampled diffraction patterns based on ranking, have now evolved and been applied both to classical decision problems and more recently to estimation problems. In addition to using these techniques for evaluating features, procedures based on Mutual Information and Rank Correlation may be used to select optimal subsets from a larger original set of features. Although the over-all feature selection problem is still basically solved by ad hoc procedures, availability of the optimization technique allows a significant amount of evaluation to be done automatically while the user concentrates on developing the specific algorithm examples illustrating the use of the diffraction pattern sampling system and pattern analysis software to solve both decision making and estimation problems are discussed.
This paper describes a system which automatically matches scene edges to a physical matrix of test edges for the purpose of estimating image quality. The system developed by EIKONIX is based upon the Itek Visual Edge Match (VEM) Station, highly modified to provide the automatic function. Random edge location and identification is performed by the operator. Scanning is accomplished by means of a one-dimensional self-scanned photodiode array coupled to the microscope by an anamorphic optical system. The data is processed internally by means of a unique eigenvector approach which yields an interesting and sensitive quality discriminator. Calibration is automatic, the machine being operated under computer control in one of several modes. Training time is minimal, and the design of the densitometric system is such that many of the common error sources and limits on dynamic range are avoided. The paper includes system design and analytical considerations as well as experimental data.
Applications of high-speed optical power spectrum analysis (OPSA) utilizing telecentric scanning systems for the automated analysis of aerial photography for several relatively simple problems are described. Cloud screening is discussed as an example where a statistical pattern recognition approach is successfully applied to OPS data. Examples of image analysis based on characterization of image structure and orientational content are presented. Projective sampling concepts, which permit image sampling based on ground coordinates in conjunction with a digital data base, are introduced. Finally, use of projective sampling with OPSA is illustrated by an image-to-image cartographic change detection experiment.
A precision continuous or incremental film transport is described, designed for use in conjunction with a laser scanning system. The transport will accommodate 70mm to 9 1/2" inch film and may be operated under manual or computer control.
In optical line-scan image generators, e.g., laser image and microdensitometer recorders, the intensity point spread function of the reconstruction or printing spot can be varied by implementation of appropriately shaped spatial filters in the optical train. The shape of the spread function can be manipulated to provide different distributions of the overlap between adjacent lines or spots. The distribution of the overlap influences both the cosmetic appearance of the imagery and the amount of resolution and contrast loss created by the blending of the intensities from adjacent lines or spots at their point of contact. In the study reported here, sampled imagery stimuli of aerial scenes displaying effects of nine different reconstruction spots were utilized in viewer experiments. Rankings of viewer preferences were generated for ideal, jitter-free reconstruction. In addition, experiments were conducted to assess performance of the reconstruction in the presence of deterministic cyclic banding and random Gaussian jitter. Emphasis of these experiments was placed on judging for information content and the maintenance of information content when spot position errors are introduced by the reconstruction.
A Hybrid Optical Correlator System for application to automated stereo compilation is described. A brief introduction to the principle of operation is given along with the description of an experimental prototype system. Test results that demonstrate the capabilities of the system are outlined.
Standard correlation is often inadequate to match two images taken with very different sensors such as panchromatic film and synthetic aperture radar. Systems requiring such images to be matched can still succeed if they work with image structure information instead of intensity information. This paper describes a technique called edge vector correlation, applies it to a real example, and shows how it can be implemented optically.
Concepts for automated pattern recognition research with hybrid optical/digital systems are discussed for application to present and future mapping and terrain intelligence tasks. Optical and digital pattern recognition approaches which may contribute to hybrid techniques are outlined and a generalized hybrid system model introduces system components, their roles and their interfaces. The automated pattern recognition research program at USAETL, which mainly involves optical power spectral analysis, is sketched for applications research, new approaches, and integration of pattern recognition systems. The paper concludes with estimates of the present state of automated hybrid optical/digital pattern recognition that point to research areas for developing viable systems of the future.