An approach to the control of multisensor image processing and recognition based on a suitable representation of control knowledge in symbolic form is presented. A hierarchical organization of control knowledge, corresponding to a decomposition of the image recognition process into subprocesses, is proposed. The knowledge for the control of the low-level and high-level phases is described in detail. The control problem involved in the automatic selection and tuning of image processing algorithms is addressed using data structures representing advised sequences of algorithms, a symbolic representation of quality control, and control strategies with backtracking capabilities. Error handling in the high-level phase is faced by a functional decomposition of the error-handling task into error states and types and by a hierarchical representation of the control knowledge for error detection and recovery. Results obtained in a real-world multisensor application are reported, and the improvement in classification accuracy obtained by the proposed error-handling mechanisms is evaluated.
The architecture of a distributed vision system is presented, with particular attention directed to the bottom-up indexing mechanism performed by a hierarchically organized network of information processing (IP) modules. Each IP module adaptively transforms input data passed by lower-level modules into more complete observations and performs a transformation that is modeled as a regularization process. This scheme is applied to the problem of recognizing objects whose 3-D shape can be described as a set of planar surfaces. Edge detection, straight-line extraction, grouping, and matching are the P modules considered. In particular, the regularization process consists of either a voting scheme or a Markov random field labeling process, depending on the level. At the higher level, a degree of belief is given about the presence of objects contained in the scene and considered in the model database. Results demonstrate both the validity of the processes applied separately at each level and the global consistency of the method.
Our objective was to develop a knowledge-based strategy for the classification, considered a cognitive process, of multisource data including remote sensing images. The main feature of our approach is the use of fuzzy sets as the representation framework. This strategy supports two supervised image classification procedures, one based on a fuzzy statistical classifier and the other on a feed-forward fuzzy trained neural network. Approximate reasoning techniques, based on fuzzy production rules, are applied to model the multifactorial evaluation process in which results from the classification of remote-sensing images are integrated with other data. An example of multisource remote-sensing data classification applied in fire prevention is presented together with numerical results and an experimental verification of the approach.
Locating moving objects in a scene is a generic task needed in numerous applications. Whenever the viewing system is static, detecting moving objects in the scene simply leads to detecting moving regions in the image plane. We describe an original framework to solve this labeling problem. The framework is based on a statistical regularization approach using spatiotemporal Markov fields. It takes temporal variations of the intensity function as observations and delivers two-symbol label maps. The solution is derived by minimizing an energy function using an iterative deterministic relaxation scheme and it is independent of the size, intensity distribution, motion magnitude, and direction of the image of the moving objects. Experiments carried out on real image sequences depicting outdoor scenes are reported. The computations are local and can be easily parallelized. This motion detection algorithm can also lead to an elementary, straightforward but useful tracking procedure applied at the moving object mask level.
We have developed a method of matching and recognizing aerial road network images based on road network models. The input is a list of line segments of an image obtained from a preprocessing stage, which is usually fragmentary and contains extraneous noisy segments. The output is the correspondences between the image line segments and model line segments. We use attributed relational graphs (ARG) to describe images and models. An ARG consists of a set of nodes, each node representing a line segment, and attributed relations between nodes. The task of matching is to find the best correspondences between the image ARG and the model ARG. The correspondences are
found using a relaxation labeling algorithm, which optimizes a criterion of similarity. The algorithm is capable of subgraph matching of an image road structure to a map road model covering an area 10 times larger than the area imaged by the sensor, provided that the image distortion due to perspective imaging geometry has been corrected during preprocessing stages. We present matching experiments and demonstrate the stability of the matching method to extraneous line segments, missing line segments, and errors in scaling.
Three-dimensional (3-D) object recognition from digitized intensity images is a central problem in providing computers with humanlike perception capabilities. We present a neural system that performs learning and classification of 3-D planar-faced objects. These objects are described through a set of line descriptors that provide a type of invariance to scaling and allow a reduction in the number of views needed to train the network. Kohonen networks have been used to allow a humanlike classification of the object views. Each network is capable of discriminating between several distinct objects and can be combined in a modular way with similar networks to build large multiobject classifiers.
We present a multiagent computer segmentation system, named KISS. This system has been implemented under our multiagent problem solver (MAPS), a generic programming environment dedicated to multiagent architecture design. MAPS entails a basic distinction between object-oriented and action-oriented agents, which allows flexible alternation between figurative and operative focusing tasks. KISS demonstrates the applicability of a multiagent approach to computer vision by offering a clear distribution of knowledge among several agents, dedicated successively to low-, intermediate-, and high-level analysis steps. Segmentation is approached through a cooperative analysis, involving both region and contour-based detection. Interpretation of patterns is made under three successive steps, using geometrical, relational, and semantic labeling, respectively. Such interpretation makes it possible to guide the selection of handling procedures to improve the initial segmentation. The potential power of such an approach is exemplified by its application to muscle fiber analysis.
The use of optical flow fields in image sequence analysis allows us to perform motion-based segmentation as well as 3-D reconstruction. Many techniques for optical flow estimation are based on some global or piecewise global smoothness assumption. Other techniques compute the flow field based only on local information. A local algorithm explicitly addressing the problem of evaluating a reliable optical flow field at motion boundaries is presented. Velocity vectors are computed as solutions of a multiwindow least-squares problem; the field is then regularized by a vector median filter. The algorithm is noniterative and nonparametric. Results on both synthetic and real-world sequences are shown; a performance comparison with two well-known techniques demonstrates the effectiveness of the algorithm in terms of noise rejection, motion boundary preservation, and speed.
We propose a model of the behavior of the thickness of the layers in photoresist Shipley 1350-J on a glass substrate. The photoresist is deposited using a centrifuge. The deposit is not uniform; the layers in the center are thicker than those at the rim. The model presented is obtained by fitting measurements to a modified Gaussian function with less than 1% error of the real behavior. The problem of uniformity is solved by diluting the photoresist with isopropyl alcohol. The proposed model uses parameters that control the uniformity of the photoresist layers on the substrate. This parameter control is important for obtaining high-quality optical holographic elements.
A method is presented for placing line, point, and ring light sources to produce uniform illumination of planar surfaces. Optimization consists in setting to zero as many terms as possible in a two-dimensional Taylor series expansion of the surface illumination. We analyze several practical lighting arrangements and discuss the significance of symmetry in lighting design. Four conditions are sufficient to produce optimal illumination: (1) place all lamps in a single plane parallel to the illuminated surface; (2) arrange linear sources in parallel pairs; (3) arrange point-source lamps to achieve fourfold symmetry with respect to any two orthogonal axes lying in the surface, this requires a minimum of four lamps; and (4) select lamp heights so that a line connecting each lamp to the center of the illuminated surface forms specific angles with the surface normal: 30 deg for line sources and approximately 39 deg for point and ring sources. Any lamp arrangement meeting these conditions produces an illumination function in which at least three orders of terms of the Taylor series expansion are zero. We also discuss circumstances under which some of these conditions can be relaxed.
In the development and analysis of sophisticated IR detection and recognition systems it is necessary to have a priori knowledge of the background clutter. The spatial power spectrum and the spatial autocorrelation functions are used to analyze the spatial structure of ground-based infrared cloudy sky images. The experimental results obtained for ground-based IR cloudy sky images do not fit the analytical model of the Wiener spectrum that is frequently used to describe natural clutter sources in the infrared. The spatial structure of ground-based cloudy sky IR images was found to be dependent on the percentage cloud cover in the image. A corrected model is developed that relates the spatial structure of cloudy sky images to the percentage cloud cover in the image.
A new method of achieving controlled high-speed infrared imaging is presented, which has made it possible to overcome some of the limitations in current imaging technology, particularly in disposal of high-rate data flow synchronized with the event or object being investigated. This technique is employed to simultaneously capture two geometrically identical images in respective spectral bands for quantitative spectrometric data reduction. The system features 12-bit data processing capability from two pieces of 64 x 128 PtSi monolithic CCD FPA operated in a noninterlaced mode at pixel clock rates of up to 10 MHz by using new electronic units developed in our laboratory. The system consists of an electronic shutter unit, a programmable external trigger controller, a DRAM memory buffer per channel, a single-channel image processor, and a PC-AT loaded with our software package. Its independently adjustable framing rate and exposure period permit up to 1800 frames/s and as short as 20 μs, respectively.
In the development and analysis of sophisticated detection systems, the ability to simulate background clutter provides a useful tool in assessing system performance. Infrared cloudy sky images can be generated using a technique based on the physical parameters of the background. Power spectra and radiance distribution functions for ground-based IR cloudy sky images in 3 to 5 μ and 8 to 12 μ spectral windows were determined experimentally in our previous research. The empirical power spectra and radiance distribution functions are used here as a basis to generate computer simulations of cloudy sky images. We require that the power spectrum and radiance distribution functions of the simulated image be in agreement with the functions that we obtained experimentally in our previous research. Realistic cloudy sky images were obtained. The simulated cloudy sky images retain the radiance distribution and power spectra functions of real cloudy sky IR images.
A photonic in-phase/quadrature (I-Q) rf phase shifter utilizing the signals from two integrated optic Mach-Zehnder modulators has been tested. This system demonstrates the capability of 360-deg phase control, less than 1-deg rf errors at constant amplitude, and less than 0.5-deg errors over a 30-dB rf amplitude taper. Using this technique, a four-channel beam-forming network was also developed and integrated with a printed patch, lightweight phased array antenna. High-accuracy linear beam control was achieved. Null patterns of more than 30 dB were produced on boresight. The null was steered to greater than 30 deg from boresight with a linear accuracy of approximately 1 deg.
The thermal change of the refractive index in YAG slabs induced by pumping is calculated under the conditions of an approximately infinite slab and one dimension when the light is transmitted along the direction . The experimental measurement is conducted by interferometry and the results measured are in good agreement with those of theoretical calculations.
An experimental evaluation of an airborne depth-sounding Lidar is described. The system, called FLASH (FOA laser airborne sounder for hydrography), is based on a scanning frequency-doubled Nd:YAG laser carried by a helicopter. An in-situ profiling instrument for measuring water parameters is also described. This system, called HOSS (hydro-optical sensor system), is also carried by a helicopter and has been used to collect data in parallel with the lidar measurements. A discussion of the lidar performance coupled to the measured water and instrumental parameters is included. Examples of measured wave forms are compared with those obtained by analytical and Monte Carlo modeling.
We report on the fabrication of novel refractive microlens arrays in photoresists, in particular on lenses with numerical apertures ranging from 0.1 to 0.3. A base layer technique is described that makes it possible to fabricate lower numerical aperture lenses in resist, compared with microlenses on glass substrates. The wave aberrations were measured in a Mach-Zehnder interferometer. Diffraction-limited performance was achieved for a numerical aperture of 0.2 and a lens diameter of 270 μm.
With mirrors of any aperture, the afocal two-mirror system has no spherical aberration. One mirror is spherical, and the other mirror is always an aspherical surface that is equidistant to the virtual parabolic mirror, the focal length of the latter being equal to the air separation between the mirrors. Therefore, a possibility exists of inspecting the aspherical surface shape by means of the known testing methods of the parabolic mirror. The system under consideration has some important technological advantages in comparison with the well-known Mersen system.
The relationship between the second-order nonlinear optical properties of a number of organic compounds and the conformations of those compounds was examined using WIZARD and MOPAC. A strong relationship was found to exist between molecular conformation and nonlinear optical response. The ratios of the calculated hyperpolarizabilities of different conformations ranged from 2.3 in the case of 4-methoxybenzaldehyde to over 240 in the case of n-hexane. This demonstrates the need to consider molecular conformation when calculating nonlinear optical effects.
A wavelet transform is used to transform data so as to improve discrimination between objects for classification. The wavelet transform is implemented in optics using the Fourier transform of the wavelets. Further, the wavelet chosen has a Fourier transform that is real and positive to avoid the use of holographic techniques. An optical setup is described in which an electronic analog sensor signal drives an acousto-optic cell that controls the intensity of an argon laser. A mechanical scanner writes the information as a line onto a spatial light rebroadcaster module containing an optical liquid crystal light valve. A lens system expands the line into a 2-D array. A real positive Fourier transform wavelet filter is placed in the Fourier transform plane of a 4-f correlator. An optical demonstration shows the formation of a wavelet transform and agreement with computer simulations. An approximation of the inverse wavelet transform is possible using only a real filter, and this is demonstrated in an optical experiment.
Ultraviolet optical properties near the edge of transparency (just above and below the band gap) of polycrystalline ALON (aluminum oxynitride, Al23O27N5) and spinel are not well characterized. The edge of transparency is commonly found to obey Urbach's rule, and this is the case for single-crystal sapphire and polycrystalline ALON and spinel as well. Room-temperature transmission and reflection measurements are made from 2500 to 1150 Å on these materials and the corresponding absorption coefficient at the band gap is represented by Urbach's rule.
This paper presents an image processing method for double-exposure speckle patterns including speckle noise filtering and fringe center-line extraction. Based on the implementation of this method and basic image processing, an automatic measurement system for the magnitude and orientation of displacement is developed.
A contouring technique, based on digital speckle pattern interferometry, is developed for investigation of surface defects in artwork deterioration studies. Skeletonized fringes are used for a quantitative automatic evaluation of the defect dimension. Some experimental tests effected on samples and ancient artworks have demonstrated the suitability of the method.
A new method to measure the surface roughness under fully developed speckle pattern illumination is proposed. This technique was developed on the basis that the roughness depends on the speckle size of doubly scattered light and that the speckle size can be determined from the second-order moment of integrated intensity over a finite area of the photodetector used. This system can determine the longitudinal roughness or the average slope of the surface profile on-line, and determine the spatial distribution of the roughness without a mechanical scanning system.
A hybrid additive-subtractive decorrelated electronic speckle pattern interferometry (ESPI) scheme using a continuous reference updating technique is presented. Unlike conventional ESPI techniques, this method uses speckle phase decorrelation between successively subtracted additive correlated speckle images, each of which contains information about the same two states of deformation of a test object undergoing vibrational stressing. It is shown that the susceptibility of this method to environmental noise caused by building vibrations or air currents is significantly lower than that of conventional subtractive ESPI methods, and fringe visibility and contrast are significantly improved over conventional additive ESPI techniques. The ability of this technique to work in a turbulent environment is demonstrated, and application to detection of defects in adhesively bonded structures, a problem of interest to the nondestructive evaluation (NDE) community, is shown.
We discuss the current status of passive millimeter-wave radiometry as a thermal imaging technique. The major problems are poor spatial resolution and slow response time. Techniques for overcoming these difficulties are identified, including the use of aperture synthesis, multichannel receivers, correlation, and inverse transform techniques. A comparison is made with infrared imaging.
In active triangulation systems, one side of the triangulation triangle is made up of a spatial light structure of well-known shape and location, causing a pattern on the surface of the target object. The other
side of the triangle is embodied by ray bundles, imaging this pattern onto a position sensor. Because nearly all surfaces show scattering characteristics between specular and diffuse reflection, the corresponding image-forming wavefronts are not of uniform amplitude, and so the irradiance of the imaging pupil is also nonuniform. If the imaging is done by an aberrated system, this can cause deviations from the image as predicted by geometric optics. An estimate of deviations resulting from defocusing by treating the imaging process in terms of scalar diffraction theory, using a linear model for nonuniform pupil irradiance is given.