A new distortion-invariant filter algorithm for object detection in SAR imagery is presented. The filters are linear combinations of eigen-images, reject clutter, have no peak constraints, and employ false-class training. The algorithm is base on two premises: filters that separate object classes are also expected to reject clutter and using eigen-data instead of actual training images and removing fixed peak constraints improve the generalization of the filter. We describe the new filter synthesis algorithm and initial test results on six classes of SAR data.
Jet Propulsion Laboratory (JPL) has developed, for the first time, a matchbox-size 512 X 512 grayscale optical correlator (GOC) with the volume of 2 inch X 2 inch X 1 inch. This compact 512 X 512 GOC consists of a pair of newly developed ferroelectric liquid crystal spatial light modulator (FLC SLM) with a 7-micrometers pixel pitch, the smallest feature size developed to date. New system architecture has been designed that has greatly simplified the system alignment and relaxed the tolerance of the Fourier transform lenses. An experimental result of automatic target recognition (ATR) applications using this GOC has been accomplished. The high-quality correlation output has validated the superior quality of the FLC SLM and new GOC architecture.
Based on the theoretical framework of the Generalized Phase Contrast (GPC) method we can design systems for a range of phase image processing applications and achieve optimal performance in terms of energy efficiency, visibility and peak irradiance. In this presentation, we give an overview of some potential applications that exploit the GPC method. These include phase-visualization and wavefront sensing, programmable optical tweezers, optical encryption systems and amplitude encoding of spatial phase patterns.
In this paper we describe a simple pose estimation based optoelectronic system for the recognition of facial images. The pose estimation described herein is a simple digital algorithm that is used to replace the composite image generation techniques while maintaining a low interclass crosscorrelation. The optoelectronic fringe adjusted joint transform correlator (FJTC) is then used to provide correlation. A description of the optoelectronic system and the entire process is presented. In addition, simulation results and comparison to SDF based recognition are provided to prove the effectiveness of the proposed system.
Recent advances in our high-speed analog liquid crystal spatial light modulators (SLMs) will be presented. These advancements include higher pixel density, smaller pixel pitch, greatly improved optical efficiency, and higher speed operation. The new VLSI SLMs can utilize ferroelectric liquid crystal (FLC) or nematic liquid crystal (NLC) to achieve phase-only, amplitude-only and phase-amplitude-coupled modulation. These devices have applications in optical processing, optical storage, holographic display and beamsteering. Design criteria and experimental data will be presented.
An incoherent correlator configuration is proposed and experimentally demonstrated that is capable of recognizing star patterns. The device may thus be employed for the orientation and navigation of a satellite or spacecraft. The correlator employs starlight directly and requires no laser or input spatial light modulator for operation. The filter is constructed form an array of mirrors that may be individually appropriately tilted so as recognize a particular star arrangement. The only other components of the system are a converging lens and CCD array detector. The device is capable of determining the pointing direction and rotation of a satellite or space vehicle. Experimental results employing the mirror array device illuminated with a point source early to simulate starlight are presented.
Optical correlation offers high speed processing capabilities of images mainly for filtering and pattern recognition applications. For a long time it has been kept in the laboratories at the state of prototype. The apparition of commercially available source of optical correlator opens the way to a wider spread of this technology. In real-time systems, illumination is an important issue, and has a strong impact on the final result of the correlation. In real-life applications, the level of illumination can vary depending of the environment, time of the day, etc. Two consequences result from the change of the illumination level, the first one is the reduction of the energy signature provided by the object, the second one is the relative augmentation of noise in the input scene. These two facts combined together can directly affect the value of the correlation peak. In off-line process these variations can be partly compensated by normalization of the input image. In live applications compensation has to be made in real-time. In this paper, results of live experiments performed with a COTS correlator are shown. The results indicate that live optical correlation can support a large amount of ambient light reduction in the input image provided that some kind of real-time feedback is provided to artificially increase the dynamical range of input gray level. The results shows that optical correlation, basically an integration operator, can identify object otherwise difficult to analyse conveniently by human being.
A new interesting hybrid digital/optical correlator, which performs the first Fourier transform electronically, hereafter referred to as a 2-f correlator, has been presented by Young. One of the advantages of this architecture compared to the classical 4-f correlator, an example of this architecture is presented by Chao, is that the optical system becomes less complex and does not require the same strict optical tolerances. The signal processing performance of these two architectures is expected to be the same if ideal Spatial Light Modulators (SLMs) with fully complex-valued coding domains are available. This study investigates the influence of a limited coding domain on the signal processing performance provided by current available SLMs. Optimal trade-off filters3 have been used for this investigation since they can be tuned, as the name suggests, to an optimal trade-off between being discriminant against distortions and being robust against noise. This has been used to evaluate the influence of different coding domains on the two optical correlator architectures. The coding domains of some commercially available SLMs have been implemented and their effect on the different correlator architectures have been analysed. These SLMs are: binary phase/amplitude SLM from Displaytech, Colorado, USA; micromirror SLM from Fraunhofer Institute, Dresden, Germany and the bipolar grey-scale FLC SLM from Boulder Nonlinear Systems, Colorado, USA. The coding domain has a strong influence on the signal processing capabilities. Generally, the 2-f architecture shows to be more discriminant and the 4-f architecture more robust to noise. The final choice between the two correlator systems depends however on available SLMs and the required trade-off between robustness and discriminance that is needed for the application.
In this paper, we propose a new image encryption and decryption system using a phase-modulated random key and the principle of interference. This optical system makes an encrypted image by the multiplication of a random phase key and a phase-modulated image. The random key and an image consist of only binary values. Thereafter a phase value of every pixel of the encrypted image is 0 or pi. The reconstructed pattern of the original image is simply performed by interfering between reference wave and a direct pixel to pixel mapping image of the encrypted image with the same random phase key. In optical experiment, both computer simulation and optical experiment show a good performance of the proposed optical method as encryption scheme.
Correlation filters using computer-generated laser radar imagery have been constructed. This paper describes how the filters were constructed and reports correlating result with the synthetic imagery used in the training set, with real ladar imagery of equivalent targets, and with real ladar imagery of false targets. A comprehensive set of images was collected on the Eglin Test Range using a direct-detect scanning ladar mounted on a 100-meter tower. Various targets were placed on a large turntable and ladar range and intensity data were collected at various aspect and depression angles. The Irma scene generation software package was then used to generate synthetic ladar imagery for these targets at a similar set of range, aspect, and depression angles. Several different techniques were used to generate the filters and to process the imagery used in this research. This paper describes one of the most successful techniques. The paper provides details on the iterative approach used to generate composite filters, describes how they were applied, and compares the results produced from synthetic and real target imagery. This experiment was considered a success since the synthetically derived filters were capable of recognizing images of real targets while rejecting false targets.
We develop a generalized minimum mean-square-error image processing filter for recognition and retrieval of noisy, blurred and obscured images. We examined the performance of this filter in four modes: (1) the well-known mean-square- error correlation filter; (2) the phase-only mean-square- error correlation filter; (3) the matched mean-square-error correlation filter, and (4) the image retrieving filter. Our simulation result show that it is possible to retrieve and recognize blurred images that are 90 percent obscured and whose signal-to-noise ratio is 0.1.
In this paper, recent technical progress in developing a compact high-speed Grayscale Optical Correlator (GOC) for real-time pattern recognition at the Jet Propulsion Laboratory (JPL) will be presented. This GOC, under partial sponsorship by the Telecommunication and Mission Operation Directorate (TMOD) program at JPL, is being investigated for spacecraft navigation applications. All up-to-date hardware development, soft simulation and experimental demonstration of real-time landmark tracking during a lander descent sequence will be reported.
We propose a class-associative correlation filter based technique for detecting a class of objects consisting of dissimilar patterns. The fringe-adjusted joint transform correlation algorithm is utilized to enhance the correlation performance, thus ensuring strong and equal correlation peak for each element of the selected class. For enhanced performance, an enhanced version of the fringe-adjusted filter is incorporated in the class-associative multiple target detection process. The feasibility of the proposed technique has been tested by computer simulation.
Correlators have been used for detecting shapes but not as often for measuring shape similarity. The complex inner product (CIP) has been used in various formulations as a shape similarity measure. The CIP is essentially a one-dimensional correlation approach to measuring similarity. One-dimensional variants of the correlation techniques including the matched filter (MF), phase-only filter (POF), and amplitude-modulated phase only filter (AMPOF) are shown to measure shape similarity in a trend that approaches human perception, however, clear performance differences are noted. The results show that the best correlator for measuring shape similarity is not the best correlator for detecting a shape. It is suggested that detection and shape similarity are fundamentally different functions that are in opposition to some degree. Ideal detection and ideal similarity measurement functions are explored. The degree to which various formulations of correlators approach the ideal functions of detection and similarity measurement are shown as well as results from human psychophysical experiments.
In this paper, our experiment consists of three steps to recognize printed music. The first step is the pre-processing stage: finding threshold for binary images, identifying staff-line parts, and removing them. The second step is the recognition stage. We first classify notes and other symbols by their sizes and characteristics. The skeleton structure analysis is used for recognizing music notes due to their complex combination of piano scores and the back-propagation and projection profile method are used for other symbols after their normalizing. The last step is the review stage. In which we investigate their syntactic validity and correct unrecognized or misconceived symbols.
Hopfield net is a typical example of a one-layer, feed-back neural network containing a layer of binary neurons and a linear feed-back matrix - the connection matrix. It was first formulated by Hopfield using nonlinear-differential equations and later by Grossberg using differential-integral equations. The nonlinear properties of the network derived form these formulations are scarce and non-systematic because of the difficulty of obtaining the complete solutions form these formulations. We use a simple discrete- algebra formulation which contains only one nonlinear operator - the threshold-logic operator, or the SGN operator. Many interesting and systematic anomalous nonlinear properties can then be derived. These properties include, eigen-state storage, associative storage, domain of attraction, content-addressable recall, fault-tolerant recall, capacity of storage, binary oscillating states, limit-cycles in the state space, and noise-sensitive input states. This paper will describe the physical origins of these anomalous nonlinear properties and the simplified mathematical analysis that leads to the derivation of these properties.
In this paper we introduce three weighting algorithms for performing shift-invariant heterogeneous phase-restricted correlation filters that are capable of identifying an object as belonging to a certain class while rejecting any object that is not a member of that class. We compare the performance of these highly discriminative filters to the performance of the phase-only filter, and the non- discriminative matched correlation filter, in similar circumstances. Even when the proposed filters achieved proper classification, the intensities of the correlation from heterogenous targets were much smaller than those from homogeneous targets. To increase the intensities of these hetero-correlation peaks relative to the autocorrelation peaks, we also introduced a fractional power law into the filter's transfer function, thereby controlling the rejection capability of the filter.
In this paper a 2D homodyne and heterodyne technique for imaging objects embedded in an opaque scattering medium is introduced. Our imaging approach is based on heterodyning of light with different Doppler shifts scattered from objects of two different textures or from an opaque object and a textured scattering medium. We report on the initial demonstration of pulling signals out of noise for an object hidden behind a scattering medium. Enhancements of signal- to-noise ratio of the order of 50 have been achieved utilizing a 2D holographic phase-sensitive detector.
Previously, most optical music recognition (OMR) systems have used the neural network, and used mainly back- propagation training method. One of the disadvantages of BP is that much time is required to train data sets. For example, when new data sets are added, all data sets have to be trained. Another disadvantage is that weighting values cannot be guaranteed as global optima after training them. It means that weighting values can fall down to local optimum solution. In this paper, we propose the new OMR method which combines the adaptive resonance theory (ART-1) with the genetic algorithms (GA). For reducing the training time, we use ART-1 which classifies several music symbols. It has another advantage to reduce the number of datasets, because classified symbols through ART-1 are used as input vectors of BP. And for guaranteeing the global optima in training data set, we use GA which is known as one of the best method for finding optimal solutions at complex problems.
Neural networks with discrete neurons cannot be studied form the differential-integral formulation because of the mathematical ill-behaviors in continuous differentiation. But they can be studied quite conveniently with discrete mathematics and N-D geometry. This paper derives the learning system of the discrete neural networks and then solve the learning problem with a mathematically very efficient scheme - the noniterative scheme using the concept of N-D geometry. The solution involves a novel approach of finding the extreme edges of a set of training pattern vectors which form a convex cone in the N-D space.
The new type of compact optical correlator, a real-time pattern recognition system, has been developed in our laboratory, which is designed by means of the shortened optical architecture and the spatial folded architecture. When Vander Lugt. Correlator is needed, we may make full use of the whole system; on the other hand, we can take advantage of the upper layer in the folded configuration system if the compact optical correlator is only used as joint transform correlator. The architecture of the optical correlator is carefully designed and fabricated, mainly including the design of two FT lenses, the making of illumination source and the choice of SLM. So the hybrid system possesses such good performances as high-speed information processing, vibration strength and small bulk. In this paper, we emphasize the application of the correlator as joint transform correlator. Input joint image is preprocessed using Roberts operator. In the Fourier plane, the differential operation is applied to the joint power spectrum (JPS). Compared with the classical joint transform correlator, the new joint transform correlator (NJTC) removes the dc component of JPS, yields sharp correlation peaks and offers high detection efficiency and discrimination ability. Meanwhile, noise-resistance is greatly enhanced in NJTC. Preliminary results are shown by computer simulations.