Computer vision has been extensively studied in Japan for advanced automation. This paper describes how Japanese R&D of the industrial vision systems have made significant progress in the last ten years, and also gives an outline of research activities in government institutes, universities, and laboratories in industry.

In the luster mensuration, luster is usually measured without differentiating its quality and quantity. The author discriminates the one from the other as Quality of luster ; the characteristics based upon the distribution of reflection from parts of surface which are very small to the extent that the resolving power of the eye comes into question. Quantity of luster ; the macroscopic characteristics based upon the average reflection from a reasonably large area of the surface which is regarded as uniform in luster. On this basis, an automatic recording "microlustermeter" has been devised to measure the quality of luster and an analytical formula is derived for the recorded curve to show the evaluation of the quality of luster. We apply this measuring method for evaluation to the metalic twinkling painted surface, the luster of the silk and cotton textiles, and the some materials have different texture.

Image processing hardware, a high-speed image processor and an intelligent image memory unit, are presented and their applications to the industrial image processing are discussed. An image processor has been realized on the basis of local parallel computation architectures, which contribute to speeding up major calculations and to reducing the hardware cost and complexity. An image memory unit has been implemented to manipulate and display the image data with a highly intelligent control, which helps man-machine interaction to be effective. Although these processors are conceptually designed for general image processing purposes, several feasibility studies have revealed that they can be utilized as com-petent simulation tools for research works on shape recognition, particle measurement, real-time signal processing, etc.

This paper describes a visual system for the purpose of pattern recognition in a group of objects, each of which is projected respectively as a circular image in a 2-dimensional picture observed through a TV camera. The recognition is based on the following. The image of an object is symmetric with respect to two lines, vertical and horizontal. The location of the object in a scene is determined by their intersecting point. The primary concern is detection of every line of symmetry. A symmetry operator and a metric. which evaluates the results of applying the operator to a binary picture are introduced for this purpose. The symmetry operator has the following advantages: 1) it simplifies the recognition algorithm; 2) it is shown that, in the experiments, it enables to get a correct re-cognition in spite of the presence of noise; 3) the function can be performed easily by a hardware system because of a local operator. An automatic handling system of the models of steel coils has been developed for an application of the visual system. The paper also presents the recognition process of steel-coil models and describes the mathematical model of the optical system for the coordinate transformation.

In this paper, a virtual system for automatic detection of fabric defects is introduced and the performance of the system is investigated experimentally from a software-oriented point of view . Major discussions on this software system are focused on how the observed noisy data can be made clear without degradation of resolution power of the image and on how several kinds of fabric defects can be detected from the noisy background. Following results are obtained experimentally: (1) Resolution power of the input data below the nominal fineness of a fabric sample(cotton) still provides approximately sufficient information of defects. (2) As a preprocessing of the fabric image data, projection procedure has efficient and appropriate properties in order to reduce background noise. (3) Combining the projection procedure with thresholding procedure in this system 'almost all kinds of defects can be detected automatically with a little misjudgement.

This is a new mask Inspection technique based on pattern width measurement that can check the fatality of defects. Mask quality depends on pattern width and defects in a fatal area where a normal pattern is seriously damage. In this technique, pattern width measurement and fatal defect detection with a laser beam are used for inspection of masks of PC boards. In order to determine the direction of measurement, diffracted laser light at the pattern edge and spatially divided photodetectors are used. The new inspection system (l) measures pattern width, (2)does not require the original data for comparison, (3)detects fatal defects, (4)has no special alignment and (5)detects defects as small as 10μm can be isolated with 100% accuracy.

The present system for identifying latent fingerprints left at the scene of a crime still depends on the skill and patience of trained fingerprint technicians because of the low quality of the latent fingerprints. An image processing & feature extraction technique has been established in order to detect minutiae (ridge endings and ridge bifurcations and ridge counts useful for latent fingerprint identification. The algorithm of the latent fingerprint identification using these features has been evaluated. Each feature is detected from the gray scale image of fingerprints and described by minutiae positions, minutiae directions and interminutia ridge counts. In the latent fingerprint identification, minutiae pairs are selected according to the similarity between latent fingerprints' features and registered fingerprints' features in a file. The similarity between two fingerprints is obtained through the comparison of selected minutiae pairs. In our experiments, we attempted the feature extraction of 500 registered fingerprints and 200 latent fingerprints, and the matching between them. The experiments were carried out using a high speed microcomputer, a flying spot scanner, a color display, and a tablet unit, which con-nected with a general purpose computer system (NEAC 2200/200).

In many production chains visual inspection of products is an important manufacturing consideration with respect to quality control. Recent progress in image processing and pattern recognition led the way to economically justified applications: modern technology enables the construction of such automata featuring high reliability and constancy. This paper describes an automatic visual inspection machine performing intelligent control tasks in very short time. The possible applications include dimension control of products, inspection of objects on shape, grey-Mess or texture, sorting of objects, positioning, etc. The design is based on a fast synthesis procedure of a TV-image by profile extraction. The features to be controlled are extracted from these profiles and compared with upper/lower limits obtained from a learning process. The system performances are expressed by the elaboration of an industrial case: the real-time visual inspection of reed switches. More than 35 features are detected and controlled for each switch within I second. Several experiments with a prototype version have proven the feasible use as an industrial on-line controller.

An earlier article described a laser flying spot scanner for investigating the surface finish on the bores of components such as the hydraulics cylinders used in motor vehicle braking systems. The quality of finish on their internal surfaces must be very high; even small scratches or blow holes may lead to failure of the braking system, with dangerous consequences. The previous paper outlined a method for processing the signal from the photodetector, so that such defects may be identified. However, difficulties were experienced with this method of signal processing, which proved to be unduly sensitive to low frequency variations in the observed intensity of light reflected from the bore. Improved signal processing was therefore required. In this article, we describe the results of an experimental investigation using an interactive image analyser. It has been possible to demonstrate, by incremental simulation, that scratches can reliably be detected using an algorithm based upon a procedure which seeks and removes intensity anomalies.

This paper reviews some of the activity in the UK directed towards realisation and exploitation of practical shop-floor automatic inspection systems. There are still considerable barriers to widespread applications because of the generally specialised nature of each situation, and the complexity and cost associated with achieving satisfactory solutions to image analysis problems. However, rugged solid-state imagers and microprocessor-based signal processing systems are beginning to advance the state-of-the-art to the point at which autonomous and robust equipment of moderate sophistication can be built at reasonable cost. The impact of microelectronics on performance and cost of equipment for inspection applications, and the problems of justifying and maintaining this equipment, are examined. Projects in which Sira Institute has been involved are referred to briefly to illustrate some of the points made.

This paper reports the development of a unique computer-controlled LLTV camera. The camera was developed specifically to meet the rigorous requirements imposed by computer tomography, computerized video processing for image recognition, and industrial X-ray inspection systems. The first large-scale application of the new camera is by researchers at the Mayo Foundation as a key component of the Dynamic Spatial Reconstructor (DSR) Program. The DSR is a medical diagnostic tool which will be capable of acquiring up to 28 X-ray images every 1/60th of a second. These images are processed by computer reconstruction algorithms to produce three-dimensional radiographic images of moving organs--images which would be unattainable with existing X-ray procedures. The camera features several unique capabilities which facilitate image acquisition and image quality. Examples of these are high-speed photocathode gating, a triggered scanning mode which holds the scanning electronics in areset condition until an image is acquired and a scan-initiate command is received from the computer, 0.25 percent absolute geometrical accuracy at any point on the scanned image, the capability to correct distortions introduced by optics or image intensifiers, and a standby mode which enables the system to be quickly powered up for intermittent duty operation.

A flexible, reliable pattern recognition system has been developed that operates under the difficult viewing conditions encountered in many industrial applications. The system is flexible because it can be trained and retrained by an unskilled operator to recognize a wide variety of pattern sets. The system is reliable because extensive preprocessing of the scanner signals is included to provide immunity to variations in lighting, surface reflectivity, pattern distortion and pattern registration. The system has a number of unique features. A different subset of the pattern set can be selected for recognition in each position in a row of characters. A varying number of characters per row can be accommodated. The patterns can be read a number of times before making a final decision. The reject level can be set by the operator. Also, the operator may overide recognition with manual input when necessary. Initial results achieved with this system indicate that practical solutions exist for the difficult pattern recognition problems that arise in an industrial environment.

The OPTOMATION TM Instrument System provides on-line capability for 100% non-contact inspection of parts or assemblies on high speed production lines. System operation is based on measurements which are made by extracting key visual features or discriminants from the image derived from a solid state Charge Injection Device (CID) camera, comparing those measurements with a set of predetermined values which establish acceptance or rejection criteria, and generating output decision signals to drive the system under control. In this paper, performance characteristics of the PN2303 OPTOMATION TM Instrument System are outlined. In addition, several specific system applications are described.

The advent and evolution of the monolithic photodiode array, in the form of the self-scanned (SSPD) and charge coupled (CCPD) photodiode array, and the microcomputer have been major component advances in electro-optical image sensor system's technology. This paper describes the basic aspects of the photodiode array, and then the concept of image sensing by line and matrix photodiode arrays-with applications emphasized. A statement on the objective of an image sensing electro-optical system is then provided, along with an identification of a system's major components, their system's function, and general applications comments on each component.

In the manufacture of automobile and truck frames varied patterns of holes of different shapes and sizes are either pierced or drilled in the frame member. Holes that are missing due to broken punches, if undetected, result in extra expense for the automotive customers on their highly automated assembly lines. Conventional means for inspecting for the presence of holes, such as discrete sensors or checking pins, have been impractical because of inflexibility to adapt to annual model changeover in addition to limitation in speed of inspection. An automatic means for 100% inspection of automobile front cross bar assemblies to determine whether the correct number of engine mount holes exist was developed at A. 0. Smith using the solid state video camera as the sensor. Video signals from a pair of cameras focussed to areas of a conveyor line where the assemblies pass are quantized to either black (0) or white (1) signal levels and fed to a specially designed logic system, which performs the following functions: -- Detects that the area of the assembly to be inspected is in the proper position in the camera field of view. -- Processes the binary image in that area by counting a topological property known as the Euler number that corresponds to the number of white "blobs" or "holes". -- Compares the resulting count to a number stored in processor memory for that part to determine acceptance or rejection of part. System has been on-line since February, 1978, operating in a harsh manufacturing environment.

We have shown that an electro-optical system can be used for 100% inspection of thick film hybrid circuits. This work was supported by the U.S. Army Electronics R&D Command under Contract No. DAAB07-77-C-0585 entitled "Automatic In-Process Microcircuit Evaluation." A high resolution (10,000 TVL/H) Return Beam Vidicon (RBV) is used to image an entire 2-inch by 2-inch substrate. The RBV is operated with computer controlled electronic steering and zoom to provide an appropriate level of detail for rapid sequential frame inspection. Video from each frame is compared with that from a referenced image stored on a video disc. Differences are displayed on a color TV monitor and processed by the computer to identify and characterize faults. Various manual and automatic inspection sequences can be programmed readily. This technique can inspect hybrid substrates at rates of 750 per hour, and its efficiency makes 100% inspection economical. This paper describes the hardware instrumentation and computer algorithms, and reports functional inspection results achieved using sample hybrid circuits with built-in flaws.

A crucial step in the recognition or location of an object in an image is the proper identification of object features. If the features are not uniquely characterized by their local appearances, as is often the case in programmable assembly, the matching technique must base its decisions on the relative structure of the features. In this paper we describe a technique that uses the relative positions and orientations of the features to determine the correspondence between features of an object model and features observed in a picture. A graph is constructed in which maximal cliques (i.e., completely connected subgraphs) represent mutually consistent assignments of model features to observed features. The technique is a robust, general-purpose way to match structures. However, in practical applications its use is restricted to moderately sized graphs because the algorithm that locates maximal cliques is apparently exponential. For tasks that require the analysis of large graphs a few techniques are presented to reformulate them so that smaller graphs are sufficient.

Kruger: I'll begin by asking the panel what and where the major focus of government-sponsored research and development in the area of industrial visual inspection and automated assembly has been and is now. I'd like to start with Dr. Brook of SIRA.

The results of an on-going research project to assess the use of resonance-neutron computed-tomography for the assay of fissile materials will be discussed. A simulation study has indicated the potential value of this technique for the nondestructive inspection of a standard to be fabricated by the National Bureau of Standards.

In industrial diagnostics, direct imaging of dynamic variables is often as important as frame sequences of configuration and composition static distributions. There have been only a few studies presented on the application of transverse axial tomography techniques to the direct development of dynamic images. In this paper, a new nonintrusive transaxial tomography method for the measurement of three-dimensional fluid velocity profiles is suggested. The technique employs the sensing of the transit time of a physical signature which is carried by the fluid. This signature is detected at locations marginal to the fluid flow region and thus usually is represented by a configuration convolution over a region of the flow. Since time of fluid transit is the variable which is imaged, the technique is labelled Transit Time Transaxial Tomography. To aid in development of the idea of the method, it is progressively detailed for the special case of a radioactive nucleus decay signature and a fluid flow meter concept which could be applied to a nuclear power plant. Generalization of the ideas are then presented in the conclusion of the work.

Through the use of dual-energy CT, it is possible to obtain a two-dimensional mapping of average electron density and average atomic number within the imaged plane. Preliminary results obtained with medical scanners indicate that these data may provide a reliable criterion for differentiating between explosives and innocent materials contained within a package or suitcase without resorting to imaging, pattern recognition, or operator interpretation. The results of a preliminary systems study indicate that a practical system based on this principle of detection is feasible. Currently, there is a program for the development of a breadboard dual-energy Cm explosives detection system tailored to meet the performance characteristics reauired for a checkpoint operational scenario.

A computerized tomography system has been developed suitable for the examination of inanimate objects of widely varying composition and shape. Photons provided by x-ray tubes and isotopic sources have been utilized to scan various industrial objects. Transmitted x rays or 15-rays have been detected with plastic scintillators operated in the photon (pulse) counting mode. Tomograms (cross-sectional density maps) have been reconstructed for a number of objects, including polypropylene cartridge filters, concrete piers, electric power poles, steel I-beams, and electronic components. The results indicate that the technique will find extensive application in quality control, in manufacturing processes, periodic inspection of structural elements subjected to deteriorating forces such as stress or environment, and non-destructive testing in general.

We describe a mathematically-rigorous easy-to-use portable topographical mapping technique that within seconds or less can determine the heights of a surface at 16,000 (128x128) distinct sample points. The necessary data derive from (narrow-optical-bandpass) TV images of a surface which has been illuminated and space-coded by a rapid sequence of laser-projected dot patterns.

Applications and system hardware are described for a new comparative topographic spatial mapping device. The device will remotely, automatically, and non-destructively measure any specified surface. A unique optical method, generates a well-defined orthogonal array of individual laser beams. Variable adjustment of beam pattern spatial frequency permits either wide-angle coverage of large-sized objects or high resolution capability for detailed areas. A novel programmable electro-optic filtering system encodes the beam array. A synchronized electro-optic selective wavelength scene capture system decodes and stores scene data in real-time in ambient light. The hardware described interfaces with suitable algorithms for the rapid calculation of the three-dimensional coordinates of sample points which mathematically define the surface of an object. The derived spatial coordinates of surface sample points then form a data base for numerically-controlled fabrication machines. A completely automated real-time scene acquisition and analysis capability could lead to machine interactive systems for pattern recognition, casting or mold comparisons, vector analysis of dimensional changes, and servo-controlled robot vision applications. The device can be ruggedly configured for use in surgical operatories, hazardous industries, outdoor inspection, etc.