In this paper a new morphological method is proposed for performing edge detection and image filtering at the same time. These operations are useful for high level image processing systems performing tasks such as pattern recognition or region segmentation. In literature two main classes of methodologies are proposed: Bayesian methods, that allow one to obtain good results with the drawback of an expensive computational load, and morphological methods that involve a better computational load and less accurate results. The proposed method is based on both mathematical morphological techniques and Markov Random Field based techniques. In this paper, experimental results are shown and it is possible to conclude that the method is suitable for processing images corrupted by structured impulsive noise such as SAR images.

THe problem of dense motion field estimation usually comes down to the minimization of a non-convex energy function. This minimization is achieved thanks to multigrid methods or more often using the first order approximation of the structural model. In this paper we propose a different approach based on the estimation of the image properties which lead to a convex hamiltonian. These properties are used to define a continuation algorithm: approximations of original images are computed using these conditions and used for the first stage of our continuation algorithm. The modeled images are then gradually transformed back to the original ones while the process evolves. This method is tested on synthetic sequences and its results are compared with those of other methods.

In this paper, we address the problem of the localization of a target with random grey levels appearing on a random background. The optimal processor for any white Gaussian statistics of the target's and the background's grey levels is the white Gaussian -statistically independent region processor. If the statistics of the grey levels of the input image are no white, it is shown that a whitening preprocessing can allow one to model with a a good approximation the statistics of the preprocessed image by three white Gaussian probability density functions, which characterize the target, the background and the boundary between the target and the background in the preprocessed image. The preprocessing can thus generalize the domain of application of the statistically independent region processors.

In this contribution, we propose an optoelectronic hit/miss morphological transform for real-time quality control by image moire analysis,that integrates a VanderLugt optical correlator and a digital signal processor associated to a vector co-processor. The procedure for real-time defect detection is a three stage process. The first step is to enhance the moire image, using the wavelet decomposition and a multiresolution approach. The second step is to automatically segment the enhanced moire image, using the moment preserving thresholding algorithm. The third step is to apply the morphological hit/miss transform to directly recognize moire images that correspond to defectuous objects. This new procedure of defect detection by global analysis of moire image data has been compared to another new technique that is based on multidimensional supervised classification of optical correlations between the test object moire image and reference moire images.

This paper deals with the problematic of the segmentation of natural images based on the fractal models. These models are based on the concept of measure of random sets and its self- similarity, and lead to the estimation of a single parameter for a natural texture: its fractal dimension. Different approaches to the implementation of the fractal geometry to the texture study are described and their properties stressed in order to obtain a close relationship between the humans point of view and the estimated fractal variables: the fractal dimension and the fractal density. The Hausdorff geometry of the measure in connection with the fractional Brownian model allowed to correlate the fractal dimension with the short range values of the autocorrelation function of properly transformed natural images, and the local definition of fractal dimensions of natural surfaces. The box counting and the covering blanket methods and algorithms were implemented and applied to estimate the fractal dimension, the lacunarity and the fractal signature of images of paper sheet and cork agglomerate surfaces. Results were statistically validated using the Kolmogorov-Smirnoff test statistics.

A variety of time delay and integration (TDI) arrays have been developed. The TDI image sensor offers significant improvement in performance over a linear CCD-sensor with respect to sensitivity. This is particularly significant in low light level operation since the effective exposure time is increased by a factor which is equal to the number of the TDI stages in the sensor. Our previous work has indicated that the TDI camera can be used with viewing angles up to 30 degrees from the surface normal. The effects caused by vibration of the viewed surface become critical when using a TDI camera at non-zero viewing angles.FOr this reason, a study was made on the vibration effects. This paper reports this study, concentrating ont he relation between the vibration of the inspected surface and the sharpness of the image. The contrast perceived by the TDI camera as a function of the vibration amplitude and frequency is measured. The measurements are performed using a special test pattern in a dynamic situation. The results based on the contrast modulation function (CMF) in the vertical and horizontal directions are reported. A hypothesis of the effect of vibration on the CMF was formulated as a rule, which was then tested with measured data. Tentative result support the hypothesis.

The use of photorefractive materials such as Bi₁₂SiO₂₀ as a dynamic holographic medium is becoming an interesting alternative to current liquid crystal based modulators in real-time optical image processing. In this paper we present an experimental realization of optical correlation for pattern recognition by means of a photorefractive joint transform correlator. The correlator operates with a liquid crystal television as the input and a photorefractive crystal at the recording plane. We deal with two possible ways to register the Fourier plane information: direct detection of the joint power spectrum or obtaining the spectrum from the phase-only distribution of the joint input. We also compare the latter case with the performance of a binary joint transform correlator. Analysis, simulations and experimental results are presented.

A new method of low contrast image (LCI) analysis is presented. The method represents a generalization of the known Zernike's method of phase-contrast imaging (ZMPCI), because it covers a wide class of LCI. From the other hand, the new method is further development of the ZMPCI, because it is based on using a new phase-space information characteristics and ensures more sensitive and detailed LCI analysis. The method is completely digital and any analog operations are not needed. It allows to obtain an effect which is equivalent to focus depth variation. From mathematical point of view, the method belongs to domain of nonlinear image processing. Results of experimental investigations are presented.

This paper presents a hierarchical method for segmenting multi-textured images. Using a pyramidal construction, texture features are extracted from each level: one based on the presence of contours and the other on a special co- occurrence matrix. Actually, this matrix makes an inventory of the occurrences son-father between two consecutive levels through the pyramid. This model is furthermore adapted for texture classification, using the intrinsic directions of the overlapping pyramidal structure. With this multiresolution scheme, the human vision is simulated in its attention focusing processes, via an individual and a contextual analysis of each textured region.

In this work we present the application of the Kolmogorov- Smirnov binary-domain measure of similarity to grey scale targets. This generalization consist of the generation of a template by thresholding the target. The threshold level is selected according to an optimality criterion with respect to a metric of performance of the detection algorithm. We examine the reduction of the computation time of the algorithm and its noise robustness, and finally we show an example of application.

In this paper, we will study the application of neural networks as filters for image processing tasks. Also, we develop a methodology to apply the neural networks as image filters. We use this methodology to develop image filters based on neural networks that we called neurofilters. We apply neurofilters to some image processing tasks such as edge detection, image restoration and image segmentation. Finally, we compare the neurofilter results with other traditional image operators.

It is presented a technique for enhancing the resolution of a 2D detector by meaning of multiple acquisitions and subsequent software processing of the acquired images, without modifying the optical configuration of the instrument nor the characteristics of the detector. The results of the application of the technique and the numerical algorithm are presented and discussed in terms of potentialities and limitations due to the intrinsic spectral broadening of the signals and to other sources of error, as uncertainty in the sub-pixel displacements or variations in intensity of the source. To check the reliability of the theoretical analysis, various software simulations have been performed, and one of these is presented in detail. Finally, some experimental results are presented, using a photon- counting multianode microchannel plate array detector system in the near ultraviolet region, with pixel size of 25 micrometers ². A toroidal mirror focuses on the detector the image of a target, mounted on two micrometric translators, which allow the 2D displacements. Images of two fine wire mesh grids are presented, with period respectively 100 micrometers and 50 micrometers , and wire thickness of 13 micrometers . The reconstructed images are good, showing the asymmetrical profiles of the grids and a very low distortion.

This paper presents a fast approach to the problem of velocity field estimation with Markov random fields. First, we propose to estimate the unknown velocity field by using a joint Markov random field through a convex markovian model which is called the energy function. Secondly, the estimated velocity field is determined explicitly by calculating the minimum of this energy function. The result obtained are compared in terms of CPU time and estimation quality to those obtained with the ICM.

In this communication we present a method to correlate a scene with a filter using a binary joint transform correlator. The method consists in defining a suitable threshold function to binarize the joint power spectrum. By suitable binarizing the joint power spectrum it is possible to obtain correlations equivalent to those obtained with spatial correlation filters with a VanderLugt correlator. We analyze the problem theoretically and also provide some simulated results.

The concept of the liquid crystal active lens basically amounts to adding a phase-only spatial light modulator to a classical lens. The modulator controls the phase of the optical wave in the pupil and achieves an arbitrary wavefront shape, that is limited only by the available modulation depth and resolution. We propose to apply the liquid crystal active lens to optical image processing. We demonstrate experimentally that directional edge extraction can be accomplished by subtracting shifted frames. The width of the edges can be varied continuously with our system. Focal length shifting can also be controlled continuously to yield low-pass filtering. After subtraction from the original frame, high-pass filtering is obtained as we show experimentally.

In this work we present a method to implement optimized binary amplitude phase only filters with spatial light modulators that produce phase only modulation. The technique consists in the addition of linear or quadratic phase codes to the phase distribution of the phase only filter in those pixels that must not blocked in the binary amplitude phase only filter. This technique permits to separate the optimized correlation. Numerical simulation and experimental results are presented.

This paper explain a new subjective evaluation method of image quality for telescope with changing contrast illuminator. Authors have designed a new illuminator equipment with changing contrast chart and tested some set of telescope samples. In the paper authors have analyzed and discussed these results. An available judge index on subjective evaluation of these systems was suggested.

We propose a bayesian approach adapted to practical target detection and location tasks where the spectral density of a Gaussian additive noise is unknown. We demonstrate that the nonlinear joint-transform correlation, which is frequency used in optical correlators, is an accurate approximation of this optimal bayesian processor. This results constitutes a theoretical support for the use of nonlinearities in optical correlators.

The domain of color camera applications is increasing all time due to recent progress in color machine vision research. Colorimetric measurement tasks are quite complex as the purpose of color measurement is to provide a quantitative evaluation of the phenomenon of colors as perceived by human vision. A proper colorimetric calibration of the color camera system is needed in order to make color a practical tool in machine vision. This paper discuses two approaches to color measurements based on a color camera and includes an overview of practical approaches to color camera calibration under unstable illumination conditions.

This paper is focused on the design of a machine vision system to solve a problem found in the manufacturing process of high quality polished porcelain tiles. This consists of sorting the tiles according to the criteria 'same appearance to the human eye' or in other words, by color and visual texture. In 1994 this problem was tackled and led to a prototype which became fully operational at production scale in a manufacturing plant, named Porcelanatto, S.A. The system has evolved and has been adapted to meet the particular needs of this manufacturing company. Among the main issues that have been improved, it is worth pointing out: (1) improvement to discern subtle variations in color or texture, which are the main features of the visual appearance; (2) inspection time reduction, as a result of algorithm optimization and the increasing computing power. Thus, 100 percent of the production can be inspected, reaching a maximum of 120 tiles/sec.; (3) adaptation to the different types and models of tiles manufactured. The tiles vary not only in their visible patterns but also in dimensions, formats, thickness and allowances. In this sense, one major problem has been reaching an optimal compromise: The system must be sensitive enough to discern subtle variations in color, but at the same time insensitive thickness variations in the tiles. The following parts have been used to build the system: RGB color line scan camera, 12 bits per channel, PCI frame grabber, PC, fiber optic based illumination and the algorithm which will be explained in section 4.

Color machine vision has shown a dynamic uptrend in use within the past few years as the introduction of new cameras and scanner technologies itself underscores. In the future, the movement from monochrome imaging to color will hasten, as machine vision system users demand more knowledge about their product stream. As color has come to the machine vision, certain requirements for the equipment used to digitize color images are needed. Color machine vision needs not only a good color separation but also a high dynamic range and a good linear response from the camera used. Good dynamic range and linear response is necessary for color machine vision. The importance of these features becomes even more important when the image is converted to another color space. There is always lost some information when converting integer data to another form. Traditionally the color image processing has been much slower technique than the gray level image processing due to the three times greater data amount per image. The same has applied for the three times more memory needed. The advancements in computers, memory and processing units has made it possible to handle even large color images today cost efficiently. In some cases he image analysis in color images can in fact even be easier and faster than with a similar gray level image because of more information per pixel. Color machine vision sets new requirements for lighting, too. High intensity and white color light is required in order to acquire good images for further image processing or analysis. New development in lighting technology is bringing eventually solutions for color imaging.

FELIX is described which represents a physical volume three-dimensional display. A modulated colored laser beam is directed via mirrors and a computer controlled x-y scanning unit towards a transparent enclosure containing a helical shaped projection screen. To describe a physical space this screen is rotated about its vertical axis so that it occupies a cylindrical volume over time. Due to the translucent property of the screen the hitting laser beam will be scattered and visible to the observer. The position of each voxel (volume pixel) is determined by the momentary location of the laser beams intersection with the rotating helix, thus providing a volumetric display medium through which scanned laser pulses are projected. The receptors in the human eye have a temporal persistence because of a mental processing delay, and this causes the eyes to fuse the light scattered from the moving two-dimensional element into a three-dimensional image. Since the images are generated within a given display space rather than on a stationary surface, they are intrinsically 3D and may be observed directly from any position. The introduced image generation technique ensures that human factors regarding depth sensation are satisfied automatically without the need for special viewing glasses to be worn by the observer. A true 3D volume display as described will complement the broad range of 3D visualization tools such as volume-rendering packages, stereoscopic and virtual reality techniques which have become widely available in recent years. Potential applications for this development range from air traffic control to various medical uses (e. g. Magnetic Resonance Imaging), entertainment and education visualization as well as imaging in the field of engineering, Computer Aided Design (CAD) and Rapid Prototyping. Keywords: 3D display, volumetric display, autostereoscopic display, laser projection display, three-dimensional imaging, spatial visualization, voxel, air traffic control, man-machine-interface

The laser cladding process consists in adding a molten blown powder to a partially melted substrate, this in order to change the substrate properties. The aim of this paper is to present different experimental set-up associated with different image processes in order to characterize the powder stream in the laser cladding process. Information such as the particle speed, the powder distribution in function of the nozzle distance are presented. The used algorithms as well as the experimental set-up are presented and detailed.

This paper is concerned with the study of fluid flows through a cylinder head of an automotive engine. The study is performed by means of flow visualization and velocity measurement, using an optical technique named particle image velocimetry. The cylinder head is a transparent model, which can be constructed from transparent resin materials by vacuum casting and/or rapid prototyping methods. The results highlight the general features of the flow, such as laminar movement, vortices, stagnation and turbulence. Flow velocities between 0.2 and 5 m/s, with a maximum error of 5 percent can be obtained.

In this paper two types of high-bandwidth receiver channels for a time-of-flight laser radar are presented. The purpose was to increase the bandwidth of the receivers compared to the previously used BW approximately 100 MHz receiver channels to improve the single-shot resolution of the laser radar. Both of the new receivers consist of a transimpedance preamplifier followed by two voltage amplifier stages, all commercial components. One of them is equipped with a prototype GaAs-MSM diode and the other utilizes a commercially available avalanche photodiode or a PIN diode as the photodetector. The performances of both amplifier channels were measured. The upper limit of the noise bandwidth f_max of receiver channel 1 was about 820 MHz and its calculated transimpedance was 150 k(Omega) . The measured noise current reduced to input was 7.4 pA/(root)Hz. Receiver channel 2 had a total f_max of 800 MHz and noise current of 7.5 pA/(root)Hz. Its calculated transimpedance was 130 k(Omega) . The performance of receiver channel 2 was measured using it as a receiver channel in a single channel laser radar. In the pulser an avalanche transistor and a SH laser operating in gain switching mode were used. The width of the optical pulse was 2.3 ns and rise time was 900 ps measured with receiver channel 2. The timing principle utilized both leading and trailing edges of the pulse. The single-shot measurement resolution was 28.5 ps with receiver channel 2 while the signal-to-noise-ratios of the start and stop pulses in the channel were 80 and 56.

Laser pulsing techniques utilizing gain switching and Q- switching operation modes of commercially available SH and DH laser diodes have been studied to realize powerful and fast laser pulses for industrial applications of pulsed time-of-flight laser range finding techniques. Pulses with the length of 100 ps to some nanoseconds and peak power level of 50 to 200 W have been aimed at. As an example, with a high speed current driver realized with a combination of an avalanche transistor matrix and a DSRD, an isolated pulse with a peak power level of 200 W and a pulse length of 100 ps was achieved from a DH laser diode operating under the gain switching region. Internal Q-switching mechanisms of SH laser diodes enable the lasers to achieve a pulse power of 100 W and a pulse width in the range of 100-200 ps.

Very often in industry and medical applications we meet tasks, where small microscopic structures have to be inspected. The task is not just related to visualization it is also necessary to inspect and measure features of an imaged object in an automated image processing based system. One newly developed handmicroscope, called MESOP, is presented that allows for a high quality optical inspection of workpieces and objects. Due to the concrete realization with a laptop based image processing and computing environment the handmicroscope can be applied in a mobile manner. It is positioned in front of the object. The optical head, supplied with telecentric optics and a small camera, captures the image, transfers it into the PCMCIA-based frame grabber. There the images are stored and further processed due to the necessity of the operator.

We present in this paper a stereovision system that has been developed in Ecole des Mines d'Albi Material Research Center laboratory for the automatic measurement of 3D deformed surfaces, in collaboration with the LAAS-CNRS laboratory. This method uses off-the-shelf lenses, CCD cameras and frame grabber, and requires that a predefined pattern be applied to the sheet surface before stamping. The system works in three steps: (i) the stereovision system is first calibrated, (ii) two images of the part to be measured are taken and the 3D coordinates are computed, (iii) the strains are calculated from these 3D coordinates.

An angular scan triangulation range imaging sensor using an integrated array of light-emitting diodes (LEDs) and a lateral-effect position-sensing detector is described in the paper. A range map is obtained by means of fast electronic scanning of the LEDs of the array. No moving parts are needed for the scanning. The sensor can be designed according to the customer needs. Two application cases of specially developed devices are studied. In one case the sensor is used for profile measurements of cylindrical objects. For a paper roll placed at a distance of about 2 m, the diameter from 600 mm to 1100 mm was estimated with the accuracy better than 2 percent. A more compact version of the sensor was manufactured and tested for metal surface orientation measurements in surface treatment robot application. The sensor was placed in the robot end effector to perform angle measurements and surface profile determination for obstacle detection and collision avoidance. The cases presented show that the customer-design approach based on geometrical modeling of the sensor sensitive volume is a good solution for designing the sensor according to specific application needs.

Smart vision systems for high performance machine vision applications are presented. The smart vision systems are based on smart vision sensors. These are programmable circuits consisting of image sensors, AD-converters, and RISC-processors integrated on the same silicon chip. The processor, being a line parallel bit-serial SIMD machine, handles both binary and grey scale information efficiently. The instruction set is specially designed for image processing tasks.

Wear and tear generate fluffiness and pills that remain in the web surface spoiling the appearance of a fabric. In quality control of textiles it is necessary to have an objective method to measure pilling that improves current methods based on visual estimations of the degree of pilling. In this work we optimize a method for piling evaluation based on image analysis that we proposed recently. The method combined operations in both the frequency and the spatial domains in order to better segment pills from the textured web background. We considered a logarithmic in base two relationship between the area of pilling and the degree of pilling based on the human perception mechanisms.

An experimental comparison of pattern classification methods in the particular case of objects lying on the seafloor in sonar imagery is carried out. The object identification technique relies on the analysis of the object cast shadow. Different kinds of geometric features are extracted such as elongation and orientation of the shadow, Fourier descriptors, and new parameters derived from the shadow profile. The performance is evaluated using two sets of data coming form synthetic sonar images differently noised. The comparison shows better performance of multi-layer perceptron especially for poorly segmented images. Finally, the performance of the system is investigated on real images.

Some regular fractal objects can be obtained from a density function, as a product of periodical components. In the optical processing the manipulation of fundamental components of the structures involving this method is very important. We applied the filter operation to each 2D components of the Sierpinski carpet using an opto-digital device and speckle modulated by Young fringes method.

Imaging spectrographs have traditionally been utilized in aerial and remote sensing applications. A novel, compact and inexpensive imaging spectrograph developed by VTT Electronics is now available. It contains a multichannel fiber optic sensor head, a dispersive prism-grating-prism (PGP) component and digital CCD matrix camera capable of area integration. In rolled steel manufacturing, a protective oil film is applied on steel to resist corrosion while in transport and storage. The main problems in the oiling machine are film thickness control and jet failures. In this application, the spectrum of fluorescence of an oil film was measured simultaneously with parallel fibers. A relatively simple calibration and analysis procedure was used to calculate the oil film thickness. On-line color control for color reproduction is essential in both consumer and industrial products. The instrument was tested and analyzed for measuring differences in color by multivariate analysis of the spectra and by color space coordinate estimation. In general, a continuous spectrum is not absolute requirement. In these two examples, filter-based measurement would probably cost less thana PGP spectrograph solution. On the other hand, by measuring the spectrum and using an advanced signal processing algorithm one production version will cover all installations in both applications. In practice, only the fiber sensor mechanics need to be modified.

A new time-of-flight-based imaging lidar was designed to carry out measurements in the 5 to 100-m range. The concept of the device, along with an illustration of the design of a time-to-digital converter for parallel time interval measurements and a presentation of our sensor breadboard is described. Test results showing sensor performance are also presented and discussed. This lidar can be used for shape and profile measurements in space and industrial applications.

We have used a modified method proposed by neiberg and Casasent to successfully classify five kinds of military vehicles. The method uses a wedge filter to achieve scale invariance, and lines in a multi-dimensional feature space correspond to each target with out-of-plane orientations over 360 degrees around a vertical axis. The images were not binarized, but were filtered in a preprocessing step to reduce aliasing. The feature vectors were normalized and orthogonalized by means of a neural network. Out-of-plane rotations of 360 degrees and scale changes of a factor of four were considered. Error-free classification was achieved.

For the survey and inspection of environmental objects, a non-tactile, robust and precise imaging of height and depth is the basis sensor technology. For visual inspection,surface classification, and documentation purposes, however, additional information concerning reflectance of measured objects is necessary. High-speed acquisition of both geometric and visual information is achieved by means of an active laser radar, supporting consistent 3D height and 2D reflectance images. The laser radar is an optical-wavelength system, and is comparable to devices built by ERIM, Odetics, and Perceptron, measuring the range between sensor and target surfaces as well as the reflectance of the target surface, which corresponds to the magnitude of the back scattered laser energy. In contrast to these range sensing devices, the laser radar under consideration is designed for high speed and precise operation in both indoor and outdoor environments, emitting a minimum of near-IR laser energy. It integrates a laser range measurement system and a mechanical deflection system for 3D environmental measurements. This paper reports on design details of the laser radar for surface inspection tasks. It outlines the performance requirements and introduces the measurement principle. The hardware design, including the main modules, such as the laser head, the high frequency unit, the laser beam deflection system, and the digital signal processing unit are discussed.the signal processing unit consists of dedicated signal processors for real-time sensor data preprocessing as well as a sensor computer for high-level image analysis and feature extraction. The paper focuses on performance data of the system, including noise, drift over time, precision, and accuracy with measurements. It discuses the influences of ambient light, surface material of the target, and ambient temperature for range accuracy and range precision. Furthermore, experimental results from inspection of buildings, monuments and industrial environments are presented. The paper concludes by summarizing results achieved in industrial environments and gives a short outlook to future work.