What readers get from a journal article is some new knowledge or new point of view; much of that is gained by browsing rather than detailed study. This usually occurs just after the journal arrives on the reader’s desk. The browsing results in considerable cross-fertilization of ideas from various sections of our field of optical science and engineering.

Model-based techniques, neural networks, and correlation pattern recognition (CPR) are three of the most popular approaches to machine vision and pattern recognition. While the former two techniques are being widely pursued by the signal and image processing community, CPR has been mostly investigated by the optical pattern recognition community. As a result, the understanding of CPR and its applications has steadily grown over the last decade

Optimum receivers for pattern recognition with nonoverlapping target and background noise are designed based on binary Bayesian hypothesis testing with unknown parameters. We overview the design of optimum receiver to detect a noisy target with unknown illumination in nonoverlapping colored background noise. Both white and colored noise are considered to model the additive noise on the target. We show that the solution for the optimum receiver, when the additive noise is white, consists of three terms. The first corresponds to the energy of the input signal that is defined as the lexicographically ordered samples of the input image within the window of the reference, the second is the square of the correlation between the input signal and the energy-normalized reference; and the third corresponds to the energy of the whitened input signal. When the additive noise is colored, the third term is the same, however, the first two terms change such that the information in the correlation matrix of the additive noise is utilized to process the input signal. We discuss two special cases: when the additive noise is zero and when the background noise is zero. We show that when the additive noise is zero, the optimum receiver is independent of the background noise statistics. For the special case of zero background noise, the optimum receiver reduces to a matched filter that estimates the illumination of the target.

A statistical analysis is provided for the properties of the recently developed maximum average correlation height (MACH) filter (Mahalanobis et al. 1994). It is shown that the MACH filter can be interpreted as an optimum filter for the detection of targets in additive noise. A rationale is given for using a popular peak-to-sidelobe ratio metric to characterize the output of the MACH filter. Finally, the performance of the MACH filter is compared to that of the matched spatial filter (MSF) in terms of the relation between the probabilities of correct detection and false alarm, which is represented as a receiver operating characteristic (ROC) curve.

We apply Bayesian parameter estimation theory to the problem of locating a known target in an image. We investigate the incorporation of prior information regarding the target location, the presence of a background, the target illumination, and the assumption of spatially disjoint target and background. We obtain the posterior probability density functions for the target location in these different cases, together with the corresponding maximum a posteriori (MAP) estimators. These results generalize and highlight some previously introduced solutions...

We address the problem of locating a target with a random correlated texture appearing on a disjoint random correlated background for tracking applications. We first review recent techniques based on SIR models and then we show how to enlarge the field of application of these approaches by preprocessing the input scenes. With numerical simulations, we analyze the probability of correct location of the different proposed algorithms for several types of correlated targets and backgrounds.

Automatic target recognition (ATR) functionality clearly relies in some way on its knowledge of what the objects and classes of interest look like. The ‘‘comparison’’ of this knowledge to the current information on a new object determines the results of the recognition process. If the new object information or the original training knowledge base is distorted by compression or other means, there is a high potential for degraded ATR performance. The effect is clear for ATR because a consistent signature, assuming of course the signature is a ‘‘good’’ one, helps the ATR, where an inconsistent signature has detrimental performance effects. Lossy compression methods vary significantly as to the type and amount of distortion they introduce. We show the distortion comparison of 15 contractor/industry methods on 8 and 12 bit data sets quantitatively over varying ranges of compression and make obvious the compression methods that distort the target spatial signature and those that tend to minimize this distortion.

Optical pattern recognition is significantly improved by exploiting the attributes of both optics and electronics in architectures containing combinations of optical systems and electronic processors. Essential interfaces for such hybrid systems are spatial light modulators and optoelectronic recording devices. These devices suffer from technological limitations that must be alleviated by suitably designed optical architectures and efficient processing algorithms. We review several correlator architectures and discuss their adaptability to hybrid systems. Due to the computational complexity encountered in such hybrid electrooptical systems, iterative optimization methods can be efficiently employed. Some of the algorithms presented are especially useful for treating real physical systems having properties that cannot be exactly defined or quantified. Algorithms that were successful in these architectures belong to two families, the basically stochastic algorithms and the projection-onto-constraint-sets algorithms. The former is particularly suitable for applications in hybrid electro-optical learning systems, while the latter is extremely powerful for the design of spatial filters with properties that are difficult to achieve by other means. A case study is given of an adaptive correlator for rotation-, scale-, and shift-invariant pattern recognition.

We investigate composite Fourier-plane nonlinear filters for distortion-invariant pattern recognition. Different types of noise such as spatially nonoverlapping color background noise, other nontarget objects, real scene noise, and additive white noise are used to investigate the performance of the filter. Computer simulation results are presented to show the performance enhancement using Fourier-plane nonlinearities in the composite filter designs. The training image sets and the test image sets are generated by in-plane and out-plane rotation of different aircraft.

We present recent advances in the development of nonlinear extensions to the minimum average correlation energy (MACE) filter. The MACE filter and its variations have been applied to the area of automatic target detection and recognition (ATD/R). Nonlinear extensions (Fisher and Principe, 1994) have been presented based on a statistical formulation of the optimization criterion, of which the linear MACE filter is a special case. The method by which nonlinear topologies can be incorporated into the filter design is reviewed. We present recent advances to this nonlinear method as well as new experimental results applying the technique to inverse synthetic aperture radar (ISAR) data. The methods described result in faster convergence times and significantly better classification performance.

We present a high-SNR approach to estimating the orientation of distorted target objects, where the distortion is out-of-plane rotation. The presence and position of the targets are detected with a bank of distortion-invariant correlation filters. The filter set we use, known as hybrid composite filters, yields complex responses at the target locations. The peak magnitude responses indicate target locations. The correlation phase angles, at the target locations, are linearly combined into complex signatures unique to a particular orientation. A maximumlikelihood M-ary classification algorithm is used to determine the most likely orientation from a finite number of orientations. In practice, we do not use all the hybrid filters and the ones we do use are optimally selected and combined for detection and discrimination. Given the target location, additional filtering can be accomplished with an inner-product operation rather than correlation. A second set of filters, optimized for angle estimation, are applied as inner products at the target locations to construct the orientation signatures. A rigorous mathematical treatment of the optimum signature detection architecture is presented and numerical simulations demonstrate the capabilities of this approach.

Classification and pose estimation of distorted input objects are considered. The feature space trajectory representation of distorted views of an object is used with a new eigenfeature space. For a distorted input object, the closest trajectory denotes the class of the input and the closest line segment on it denotes its pose. If an input point is too far from a trajectory, it is rejected as clutter. New methods for selecting Fukunaga-Koontz discriminant vectors, the number of dominant eigenvectors per class and for determining training, and test set compatibility are presented.

New variations of minimum noise and correlation energy (MINACE) filters are considered for detection and recognition of objects and rejection of clutter in synthetic aperture radar (SAR) data. We present initial very attractive results that aid different modules in an SAR automatic target recognition (ATR) processor. For four class test set data, we show detection performance of PD>99.8% with PFA ?0.026/km2; for classification we show PC>96.6% with PFA ?0.026/km2. Selected tests with noise and obscurations present and of several objects in large real SAR background are also included. All results used full shift-invariance tests.

We describe an optoelectronic information processing system that is capable of real-time vehicle navigation and target acquisition. The system uses a holographic database, based on the DuPont HRF-150 photopolymer, to perform the desired tasks. The architecture and the performance of the system are discussed in detail.

Many image processing applications require a small, lowpower, low-cost pattern recognition system that is capable of locating and identifying objects. Miniature optical correlators can perform 2-D pattern recognition at greater rates than digital platforms of equivalent size, power and/or weight. The patented miniature ruggedized optical correlator (MROC™) module is built to meet the environmental, size, power, and weight requirements of military and rugged commercial applications, and at a cost that permits wide deployment of the capability. The second version of the MROC module consists of a ferroelectric liquid crystal (FLC) device in the input plane for high light efficiency and incorporates a reflective magnet-optic spatial light modulator (RMOSLM™) device in the filter plane for very high speed operation. The correlator has a volume of approximately 20 in.3. MROC II breadboard tests demonstrated excellent correlation peaks, i.e., excellent discrimination (SNR>25), at pattern matching rates of 1920/s on images of military vehicles.

An experimental implementation of a joint transform correlator based on a nondegenerate four-wave mixing interaction in the photorefractive material bismuth silicon oxide is described. An investigation is made of the pattern recognition performance of the correlator by assessing its ability to maintain a correlation response despite orientation changes of a model road vehicle while simultaneously discriminating against a similar vehicle. The full-bandwidth performance, essentially implementing an inverse filtering operation, was found to give good discrimination but to be highly sensitive to changes of the vehicle orientation from that of the reference function. A bandpass filtering strategy, based on the difference of Gaussian wavelet filter, was implemented in the correlator by exploiting the Gaussian amplitude profile of the TEM00 mode of the helium neon readout laser. An extensive experimental evaluation of the performance of the modified filtering scheme is presented, demonstrating considerably enhanced intraclass distortion tolerance while maintaining good interclass discrimination ability.

The robustness of the joint transform correlator (JTC) versus the VanderLugt correlator (VLC) are evaluated based on the transversal and the longitudinal misalignment. We show that the VLC is very sensitive to the transversal misalignment, while the JTC is not. Although the VLC performs better in terms of longitudinal alignment, the tolerable misalignment in recording the joint transform power spectrum (JTPS) is in the order of a millimeter, which can be easily adjusted in practice. In short, we conclude that the JTC is far more robust than the VLC.

The lensless joint transform correlator optically determines the relative x -y position of two pinholes. The pinholes are placed in parallel arms of a Mach-Zehnder interferometer. The Fraunhaufer diffraction patterns combine to form a joint transform plane. Because of the interference of the two signals, a Fresnel zone is formed in the joint Fraunhaufer diffraction pattern. This is captured and used in a second stage, just as the joint power spectrum is in a joint transform correlator. The joint transform is illuminated with a plane wave, and the output taken in the focal plane of the Fresnel zone. As with the single-lens joint transform correlator, the lensless joint transform correlator exhibits magnification between the input displacement and the output peak displacement. By using the third or fifth order focus of the Fresnel zone, the output resolution can be improved by a factor of three or five in both the x and y dimensions. The output has very good peak-to-noise ratio and position resolution. A peak-to-noise ratio of 106 dB with a resolution of 1.4 µm are observed when using the fifth order focus of the Fresnel zone.

Power spectrum subtraction is introduced into the joint wavelet transform correlator to improve its performance. Because the joint wavelet transform correlator is not a matched recognition, its performance is quite poor when the conventional joint-transform correlator is employed. By subtracting the power spectra of the target image and the reference image, the power spectrum subtraction technique may greatly improve the joint wavelet transform correlator. In comparison with the conventional joint wavelet transform correlator, the new correlator has narrower dc output, bigger space-bandwidth product, stronger wavelet transform output, and greater background-noise resistance. Both theoretical analysis and computer simulation verify its efficiency.

The design of Lyot-stop systems in earth limb viewing radiometers is first reviewed, with particular reference to the stray-light analysis methods used for instrument design and performance evaluation. These methods are given in order of increasing sophistication, up to the most powerful technique for diffraction analysis, that of generalized beam propagation analysis. The typical design features of radiometers for the limb-viewing application are discussed to summarize the important stray-light issues, including scatter, ghosting, and diffraction effects. How for multichannel instruments based on arrays of detectors, the interplay of these issues affects the Lyot-stop system is shown, pushing the design toward nonstandard geometries for which the generalized analysis method becomes essential. An example is given of the use of the technique in a system requiring a nonstandard Lyot-stop geometry, including the computation of the full diffracted energy patterns at each stage of the Lyot-stop system, illustrating the physics of the diffraction suppression, and giving insight into the design trade-offs; in particular, that between signal throughput and diffraction rejection. For one choice of this trade-off, the net diffraction response is calculated. This quantity is an important input to the ground characterization of the predicted performance.

The performance of a self-referenced two-fiber sensor system is analyzed. In this sensor, two multimode fibers of the same kind are employed for microdisplacement measurements. The analytical model is based on Fresnel and Fraunhofer approximations. Computer simulation results are also provided.

A previous electric current sensing scheme employing two metal-coated single-mode optical fibers is modified for the purpose of much wider band measurement. As the same optical setup is used, the proposed sensing scheme incorporates the use of a differentiator that differentiates the measurand before it splits into two parts and flows on the two metallic coatings, respectively. Two kinds of optical fibers (which have different cladding sizes) are tested. With regular polarizationmaintaining fibers (which have a 125 µm cladding diameter), a measurement bandwidth of ~500 Hz can be obtained. With thinner fibers (with 80 mm claddings), the measurement bandwidth can reach ~1.1 kHz. The measurement scheme proposed compares much more favorably to the previously reported sensing scheme, in which the measurement bandwidth is only 0.06 Hz.

The problem of reconstructing the support of an imaged object from the support of its autocorrelation is addressed within the framework of genetic algorithms. First, we propose a method of coding binary sets into chromosomes that is both efficient and general, producing reasonably short chromosomes and being able to represent convex objects, as well as some non-convex and even clustered ones. Furthermore, in order to compensate for the computational costs normally incurred when genetic algorithms are applied, a novel multiresolution version of the algorithm was introduced and tested. The multiresolution genetic algorithm consists of a superposition of multiple algorithms evolving at different resolutions, sequentially. Upon occurrence of some convergence criteria at the current scale, the genetic population was mapped at a superior scale by a coarse-to-fine mapping that preserved the progress registered previously. This mapping is implemented in a genetic algorithm framework by a new genetic operator called cloning. A number of experiments of object support reconstruction were performed and the best results from different genetic generations were depicted in chronological sequence. While both versions of genetic algorithms achieved good results, the multiresolution approach was also able to substantially improve the convergence speed of the process. The effectiveness of the method can be extended even further if a parallel implementation of the genetic algorithm is employed. Finally, alternate coding methods could be readily used in both the standard and the multiresolution approaches, with no need for further adaptations of the basic structure of the genetic algorithm.

To deal with such a problem as object recognition of position, scale, and rotation invariance (PSRI), we utilize some PSRI properties of images obtained from objects, for example, the centroid of the image. The corresponding position of the centroid to the boundary of the image is invariant is spite of rotation, scale, and translation of the image. To obtain the information of the image, we use the technique similar to Radon transform, called the oriented-polar representation of a 2-D image. In this representation, two specific points, the centroid and the weighted mean point, are selected to form an initial ray, then the image is sampled with N angularly equispaced rays departing from the initial rays. Each ray contains a number of intersections and the distance information obtained from the centroid to the intersections. The shape recognition algorithm is based on the least total error of these two items of information. Together with a simple noise removal and a typical backpropagation neural network, this algorithm is simple, but the PSRI is achieved with a high recognition rate.

We propose a method for line extraction in complex images. We consider a two-stage process. First, the initial measurements at each image point are given by combining the image convolution outputs obtained from directional filters. Second, a postprocessing stage, based on probabilistic relaxation, refines the initial labeling using contextual information. The major differences from other similar approaches are the explicit inclusion of the relative measurements between pixels and the careful modeling of the probability density functions of these measurements.

A novel spatiotemporal segmentation technique is further developed for extracting uncovered background and moving objects from the image sequences, then the following motion estimation is performed only on the regions corresponding to moving objects. The frame difference contrast (FCON) and local variance contrast (LCON), which are related to the temporal and spatial homogeneity of the image sequence, are selected to form the 2-D spatiotemporal entropy. Then the spatial segmentation threshold is determined by maximizing the 2-D spatiotemporal entropy, and the temporal segmentation point is selected to minimize the complexity measure for image sequence coding. Since both temporal and spatial correlation of an image sequence are exploited, this proposed spatiotemporal segmentation technique can further be used to determine the positions of reference frames adaptively, hence resulting in a low bit rate. Experimental results show that this segmentation-based coding scheme is more efficient than usual fixed-size coding algorithms.

An optical system for a high-resolution short-wave IR (SWIR) spectrograph to be employed in remote sensing applications is described. The instrument covers the spectral range from 1200 to 2400 nm and images a 9.4-deg line object onto a 2-D PtSi sensor. Spatial and spectral resolution requirements were set at 0.33 mrad and 10 nm, respectively. While signal strength calculations indicated that an aperture equivalent to f /1.8 would be adequate, because of the anticipated lower signal levels at the long-wavelength end, special attention was paid in the antireflection coating design stage to partially compensate for the large discrepancy in signal strengths found across the spectral range.

A holographic interferometer is used to measure diffusion coefficients of binary molten salt mixtures. The composition of such transparent mixtures is proportional to the refractive index. The composition changes cause local variations of the refractive index during diffusion. These variations are visualized by interference fringes, which are photographed by a CCD camera. Exact detection of the intensity extreme values of the fringes is essential to calculate diffusion coefficients. As irregular illumination and noise disturb the detection, adaptive filter algorithms are applied to reduce noise sufficiently while maintaining all relevant information.

There are some optical arrangements that closely resemble interferometric configurations but do not produce fringe patterns. An important characteristic of these devices is that they do not have two different outputs with complementary patterns, but only one. Some of these interferometers are described here, pointing out their common properties and differences. It is shown that if we open the second output, both complementary patterns will appear.

The use of a Twyman-Green interferometer to measure the wedge angles and pyramidal error of high-precision optical prisms is described. The method allows these parameters to be determined independently or in combination. Prism angles are measured with the aid of a reference flat attached to a goniometer mounting. The pyramidal error is obtained by measuring the period of interferogram fringe pattern. Prism angles up to 25 deg have been measured with a standard deviation of 3 arcsec.

We make a geometrical analysis of optical arrays having the spatial integration feature. A complete study of the ray tracing and the energy propagation along the thick lens array is made. This analysis enables us to find some design criteria for the choice of the free parameters to optimize the total amount of energy delivered on the synthetic image plane.

The optical principle of a high-speed three-dimensional scanner that can operate at high and medium resolution was developed on the basis of confocal microscopy. It was shown that 3-D detection of surfaces, essentially independent of surface characteristics such as inclination and reflectivity, is feasible. Shading effects were found to be negligible. A data rate of 2 Mpixel/s and a 3-D resolution of 5 to 15 µm at 150-mm to 1.5-mm height range were achieved. In a factory a highresolution scanner has been operating successfully for automatic solder joint inspection of multichip modules since 1994. Also, a mediumresolution system for solder joint inspection of surface-mounted devices was developed and evaluated. In view of their high inspection reliability, the developed 3-D inspection systems promise to be a powerful tool for use in production lines.

Artificial vision is an efficient means of assuring the quality of a certain class of products. The vision system must respect the industrial constraints, in particular, the production rate. The geometrical features of flaws are pertinent information used for the acceptance of the controlled product. This article presents a real-time algorithm for the geometrical characterization of defects located on warped objects. The algorithms described enable the characterization of defects by their size and their 2-D shape. Both parameters are calculated in real time by simple reference to a look-up table. The 2-D shape is obtained by a geometrical transform and an interpolation. The efficiency of the characterization method is experimentally evaluated.

Eigenmode expansion is used to find the transmission, reflection, and radiation from a step junction in dielectric slab waveguide. Although the eigenmode expansion is applied to a slab waveguide enclosed by metallic box, it is shown that from the amplitude of the ‘‘radiation-like’’ modes of this waveguide the radiation characteristics of the open waveguide can be found successfully. Several examples are presented to confirm the validity of this procedure.

Concentric spherical photoelectron diode structures can be used to generate high current densities when illuminated with intense laser beams. For applications in electron microscopy and for beam source applications, trajectory aberrations of the emitted particles are important. This paper describes Monte Carlo estimations of trajectory fluctuations produced by discrete-charge interactions in a laser-driven spherical photoemission structure. The trajectory of one photoelectron traveling in the presence of 125 randomly positioned neighbors is computed, and its apparent source position is determined as it arrives at the anode. Hundreds of such passes are averaged to estimate the aberration. Apparent source blur sizes range between 1 nm and 3.5 µm, for emitter radii between 160 µm and 2.82 mm, as the current density increases from 1 to 1000 A/cm2. Such small aberrations imply that discrete space charge effects do not significantly degrade the performance of these electron sources under most conditions.

The radiant statistics of thermal images of desert terrain scenes and their temporal behavior have been fully understood and well modeled. Unlike desert scenes, most natural terrestrial landscapes contain vegetative objects. A plant is a living object that regulates its temperature through evapotranspiration of leaf stomata, and plant interaction with the outside world is influenced by its physiological processes. Therefore, the heat balance equation for a vegetative object differs from that for an inorganic surface element. Despite this difficulty, plants can be incorporated into the desert surface model when an effective heat conduction parameter is associated with vegetation. Due to evapotranspiration, the effective heat conduction of plants during daytime is much higher than at night. As a result, plants (mainly trees and bushes) are usually the coldest objects in the scene in the daytime while they are not necessarily the warmest objects at night. The parameterization of vegetative objects in terms of effective heat conduction enables the extension of the desert terrain model for scenes with sparse vegetation and the estimation of their radiant statistics and their diurnal behavior. The effective heat conduction image can serve as a tool for vegetation type classification and assessment of the dominant physical process that determinate thermal image properties.

Current narrow-band optical filters for professional solar astronomy applications are primarily based on Lyot filters.1 However, the Lyot filter is severely limited in operation at UV wavelengths because of the lack of suitable UV hightransmission polarizers. In addition, the Lyot filter is comprised of over 50 elements ~over 30 cm long and weighing several kilograms! that require precise alignment and temperature stabilization for all of the elements.

Current narrow-band optical filters for professional solar astronomy applications are primarily based on Lyot filters.1 However, the Lyot filter is severely limited in operation at UV wavelengths because of the lack of suitable UV hightransmission polarizers. In addition, the Lyot filter is comprised of over 50 elements ~over 30 cm long and weighing several kilograms! that require precise alignment and temperature stabilization for all of the elements.

The Handbook of Optomechanical Engineering is intended to be a self-contained book covering all major aspects of optomechanical engineering from conceptual design to fabrication and integration of complex optical systems. The intended readers are those optical and optomechanical engineers and scientists who engage in the design, development, and integration of modern optical systems for commercial, space, and military applications. After going through the handbook, I find that these objectives are well achieved.