Automatic inspection allows the continuity of the quality of textiles manufacturing, and the optimization of the pinking according to the situation of the defects. A method for defect detection based on a periodicity breaking of the material densitometrical profile is proposed. After the explanation of the method, some important points like texture suppression, image block size choice, and defect characterization are detailed. Finally, the system configuration and the results obtained from the detection stage are introduced.

A self-reliance technique is developed for detecting defects in repeated pattern wafers and masks with low-order distortions. If the patterns are located on a perfect rectangular grid, it is possible to estimate the period of repeated patterns in both directions, and then produce a defect-free reference image for making comparison with the actual image. But in some applications, the repeated patterns are somehow shifted from their desired position on a rectangular grid, and the aforementioned algorithm cannot be directly applied. In these situations, to produce a defect-free reference image and locate the defected cells, it is necessary to estimate the amount of misalignment of each cell beforehand. The proposed technique first estimates the misalignment of repeated patterns in each row and column. After estimating the location of all cells in the image, a defect-free reference image is generated by averaging over all the cells and is compared with the input image to localize the possible defects.

This paper describes a system developed for segmenting multiband grayscale images into n-class labeled images at high-throughput rates. This system, which we refer to as the segmentation engine, performs supervised image segmentation using algorithms based on the statistical pattern recognition paradigm. So-called 'features' are computed for each pixel and the feature vector thus formed is presented to a statistical classifier, which uses feature information to determine the most probable class of the pixel. Algorithms are described for the following: features, automatic feature selection, classification and classifier training. While this paper describes the entire system, the algorithmic approach will be emphasized.

We introduce the Advanced Thin Film Inspection System (ATFIS), a state-of-the-art automatic visual inspection (AVI) system developed by IBM that integrates sophisticated subsystems to inspect advanced multilayer electronic packaging. The image acquisition and 'image segmentation' (material discrimination) is based on a high-radiance illumination, multiband line scan cameras and the Low-level Image Segmentation Architecture (LISA). LISA employs a decision-theoretic approach that classifies pixels according to a decision function defined over a multidimensional 'feature space' (for example, local or global image statistics). The image processing and pattern analysis is based on the Parallel Image Processing System for Inspection (PIPSI) architecture. PIPSI's key features include exploitation of emerging hardware and software technologies; highly programmable to adapt quickly to changing functional requirements; and exploitation of image and operator parallelism, making it highly scalable to meet manufacturing requirements. The pattern analysis algorithm is a synthesis of reference-based comparison and design-rule analysis.

Learning algorithms are introduced for use in the inspection of cross-sectional X-ray images of solder joints. These learning algorithms improve measurement accuracy by accounting for localized shading effects that can occur when inspecting double- sided printed circuit board assemblies. Two specific examples are discussed. The first is an algorithm for detection of solder short defects. The second algorithm utilizes learning to generate more accurate statistical process control measurements.

This paper describes work being carried out in the area of object recognition. The research is part of a project, at Kingston University (London), to automatically inspect surface-mounted assemblies. The system is designed to use some of the significant properties of the object-oriented paradigm to build a flexible architecture that can be applied to many inspection problems. The recognition system operates on a new information representation constructed from three inter-linked data structures. The application of the proposed object-oriented recognition framework, to inspect surface-mounted assemblies, will be described in this paper.

This paper describes a system known as 'The Disaster Detector', for automatic inspection of the air-bearing surface of disk sliders (disk read/write heads). It inspects for certain types of defects that are global or systematic in the sense that, when they occur, they occur on every slider in a row or, in some cases, on every slider in the entire carrier. The inspection system is described and the associated image-analysis algorithms are described in detail. The system uses standard microscope optics, a color CCD camera, computer-controlled state, laser autofocus, a video digitizer and a PC/AT (486-based).

This paper describes an image-based inspection system for surface defects, implemented on a transputer system with a high- performance transfer bus to provide fast access to large blocks of data. Data-level parallelism (the image pixel data is partitioned horizontally into slices) and task-level parallelism (the algorithms themselves can be parallelized) are utilized. Surface defects and anomalies are detected by a texture-based segmentation procedure. In the training phase the objects of interest are marked and all feature vectors implemented in the system are computed. The system uses simple statistical features, features calculated from the cooccurrence matrix, features based on texture spectrum and the fractal dimension. The time critical run phase is realized on a parallel computer. The implementation is done fully in software, allowing flexibility in the use of features and window sizes for pixel descriptors and freedom in the use of various classifier-methods. The processing speed of the segmentation system is easily scalable with the number of processing units. The performance of the system is demonstrated on an industrial visual inspection task, the recognition of surface defects of aluminium cast workpieces, where a connectionist classifier is used for pixel classification.

A round florescent bulb placed around the lens of a CCD camera illuminates a group of carbon resistors. The pictures are passed to a series of feature detectors the outputs of which are sent to a back propagation neural network. Two identical boards were each made with 112 resistors of various colors. Resistors on one board were burned. The network was trained with a test set of about half the good and bad resistors. Results showed 93.8 percent correct when recalling on the untrained half.

A new procedure is presented to introduce machine vision into a rolling mill environment for real-time on-line metal bar and rod inspection for flaw detection and characterization. Results are presented relating machine vision metal flaw signatures to flaw type and characteristics.

The new pattern inspection algorithm we developed detects fatal defects on printed wiring boards. The algorithm determines whether patterns identified by an automated pattern inspection (AOI) system are actually defective by considering the electrical malfunction that the defect will cause. A macroscopic model based on the pattern design rules and their tolerances to pattern violations is needed to evaluate defects. The algorithm classifies features around a defective pattern into 50 categories and compares the defect distribution with preset check rules. The automated optical verification system we developed captures pattern images with a CCD camera and uses verification software to evaluate defects. The process takes 10 seconds per image. We tested the system on the factory floor, and it detected all defects with less than 4.8 percent of false alarms.

Automated ceramic tableware inspection displays many of the problems associated with automated inspection. A large number of possible faults must be considered, further complicated by the large variety of patterns with which the tableware is decorated. A hierarchical approach is taken in this work involving several levels of feature extraction and classification. Faults are detected, evaluated and the item as a whole is classified as a function of the combined fault measurements. Faults can be categorized as visual, surface, and structural. This paper concentrates on one visual fault known as shading variation, a result of color pigments registering as light, dark, or irregular. This fault is evaluated by analysis of the image intensity histogram. The histogram displays characteristics which vary with shading. Deviations from the standard are evaluated and the results passed onto a further classification stage which takes results from other feature extraction stages to give an overall evaluation of the item under inspection.

This paper presents an automated system for quality control of fabrics. It is based on a linear vision machine designed to be integrated on to weavers. The aim is to detect and identify, in real time, faults in the fabric in order to perform on-line diagnosis of the weavers. A solution based on a high-resolution linear multisensor device is proposed in order to cope with the small size of the flaws which occur on wide fabrics. A front processor integrated on the acquisition board performs local operations on the successive line images as they are captured by the multisensor device. Flaws are discriminated using a set of local detection operators and adaptive models. These models are used to extrapolate the image of the fabric, so that any flaw can be detected by comparing the actual image to this theoretical model. The detected flaws are identified by means of morphological filtering and a syntactic classification scheme that discriminates the flaws.

In the glass bottle production, the detection of defects on the finish is one of the most important steps in quality control and a difficult inspection task. A new finish check inspection method is proposed, based on computer vision. A matrix camera is used to capture the image of the finish. A new image analysis approach, based on a directional gradient calculus method, is proposed, which discriminates the defects from the noise produced by the reflection of the finish. Laboratory results show the high performance of the system which improves the efficiency of the inspection and increases considerably the speed of the detection.

The surfaces generated by different manufacturing processes like shaping, grinding, spark erosion, and casting result in textures characteristic to the machining process. Investigations have been carried out for finding the suitability of texture in segmentation in the case of occluded object scene. The specimens are prepared under controlled conditions and their digitized pictures are recorded with suitable lighting arrangement. A number of texture parameters based on first-order and second- order statistics are evaluated from these digitized images along four directions. Optimum features were selected based on interclass and intraclass scatters. Two Bayesian classifiers and one linear discriminant space classifier were designed and their success rates in identifying the manufacturing process are reported.

Machine vision and automatic surface inspection has been an active field of research during the last few years. However, very little research has been contributed to the area of defect detection in textured images, especially for the case of random textures. In this paper, we propose a novel algorithm that uses color and texture information to solve the problem. A new color clustering scheme based on human color perception is developed. No a priori knowledge regarding the actual number of colors associated with the color image is required. With this algorithm, very promising results are obtained on defect detection in random textured images and in particular, granite images.

The realization of an integrated, flexible, and robust CIM vision system, suitable for performing quality-assurance surface inspections is discussed. The optimized combination of advanced optics, optomechanics, and flexible image sensor realizes a high 'virtual resolution' without penalizing the pixel transfer rate. High computation rates are obtained by complementing the fractal inspection algorithm with a dynamic hologram, a modular data flow processor, and the system computer. The integrated vision system is validated for the surface quality inspection of concrete tiles in an industrial environment. The overall system performance is discussed in detail and the potential of the system for other application fields will be addressed.

This paper concentrates on the problem of automating inspection of nonwoven materials produced in a form of continuous rolls. This paper analyzes the problem of designing and validating appropriate digital image processing techniques for texture characterization that correlate to grading by inspectors. The design involves two sub-systems: texture characterization and measurements fusion. A case study of 15 nonwoven samples graded by inspectors is used to illustrate the roles of the subsystems. The class of materials considered is characterized by complex texture patterns formed by randomly distributed fibers. The paper concentrates on the first subsystem and considers texture characterization using measurements derived in the spatial and transform domains. The specific algorithms are outlined and the results of the case study are detailed.

Inspection of loaded printed circuit boards (PCBs) using 3D laser striping range data is described. The work presented here is intended to assess the feasibility of using 3D depth data for robust identification of PCB defects. A knowledge-based control system interrogates a 2D CAD database for PCB layout information. Test strategies are then implemented with efficient multiresolution data capture. Inspection algorithms use 3D scene models to detect and interpret faults. This paper describes the implementation of the 3D range scanner, inspection tasks, control system, and 3D scene modeling. Surface-based approaches for the inspection of 3D shape characteristics have been used extensively throughout this work. This approach utilizes the inherent 3D shape characteristics of depth data. Surface segmentation algorithms are used for the implementation of board registration, testing of components, and coplanarity of ICs.

This paper describes research into a noncontact system for the measurement of components and solder on printed circuit boards (PCBs). The objective of the system is the inspection of the PCB to identify faults in the manufacturing process. A fringe pattern is generated on the PCB, using either interferometric methods or projection of a grating, and the image of the fringe pattern is captured and stored on a frame store. The pattern is then processed to reconstruct the 3D shape of the PCB surface from which parameters which relate to PCB integrity can be determined. Results demonstrate the feasibility of the method.

A new active mode technique for the measurement of 3D objects was developed using a CCD line sensor and a galvanomirror scanner. The CCD line sensor is placed in parallel to the laser light scanning plane. When a series of equally spaced laser light spots is projected on to a surface, the spot density on the surface is altered depending on the shape of the surface. By measuring the displacement of a light spot the distance between the camera and the object can be determined using trigonometry. During the measurements, laser light was vertically scanned and object was horizontally moved. The measurement error for an object of 1 m apart from the camera is 0.08 percent in scanning direction and 0.2 percent in object-height direction. A modified experiment also was performed for the purpose of shape recognition of a given object. The success of this experiment is demonstrated by using objects of rectangular, triangular and round shapes.

We present an efficient technique that uses bidirectional template matching for gross defect detection in 3D objects. The template matching utilizes computer-aided design models and is carried out on 3D data formed by merging range data collected from multiple known viewpoints. The inspection scheme assumes that the identity of the object is known but that the object's position is unknown, although constrained. A matching technique based on the Hausdorff distance is used to determine the pose of the object. The inspection scheme is tested on real range images of several iron castings.

Depth and motion measurements from images have a variety of industrial and non-industrial applications. Underlying these two measurement modalities is the ability to accurately and reliably correlate portions of images separated in space, as in the case of a stereo pair, or time, as in the case of a motion sequence. We have developed a unique correlation algorithm as well as special hardware accelerators to allow it to operate at video frame rate. The ability to make measurements in real time is invaluable in observing algorithm performance on dynamic scenes; real-time feedback has also allowed us to apply these low-level measurements in unique ways to several perceptual tasks. In this paper, we describe our progress in the areas of figure-ground separation and active object tracking using our correlation techniques. We illustrate the use of these capabilities to drive an active camera head and a robot vehicle in person-following demonstrations.

This article describes three robotic vision applications in power plants. These applications have been developed by CISE and ENEL during the last years, and will be presented in chronological order. The first one is the automatic guidance of a gripper to center on a spent fuel rod head inside the decay pool of a nuclear power plant. The second one deals with the vision system of an autonomous robot for inspection and fault diagnosis inside the electrostatic precipitator of a thermal power plant. The third one is concerned with the volume measurement of coal heaps in the coal storage park of a power plant.

A method is presented for continuous focusing of moving objects. It is based on DFD1F which is a method for determining distance of stationary objects using image defocus information. The proposed method requires recording of two images of a moving object with different degrees of blur. The change in blur is caused by varying camera parameters such as lens position, focal length, and aperture diameter of the recording camera. The two images must be recorded simultaneously in a short time period. A new camera structure is proposed for such recording of the images. In the proposed method, the requirement of a large memory space has been avoided for storing the MTF data of the camera's optical system. This is achieved by using a parameterization scheme for the MTF data. The method has been implemented on an actual camera system. Experimental results on this system indicate that the method yields an RMS error in focusing of about 4.3 percent in lens position. The image blur caused by a focusing error of this magnitude is barely noticeable by humans. Therefore, in addition to machine vision, the method has practical applications in video cameras such as camcorders.

A prototype of an image processing system for the monitoring of conveyor belt systems in video real time is described. To avoid completely hazardous situations, special requirements on the robustness and the computational efficiency of the system under heavy climatic and illumination conditions have to be met. For the computation of the belt loading and the position of the conveyed masses the profile of the loaded belt is subtracted from a stored profile of the empty belt. Both empty and loaded belt profiles are estimated using the triangulation principle. A model based algorithm combined with a two-step classificator is applied to detect large stones or other bulky objects from a sequence of profiles. The gray level characteristics of the conveyed masses are computed, using features derived from the ID histogram. The structure of the masses is evaluated using a modified fractal operator. One class out of a predefined set of classes is selected. Image processing hardware based on the digital signal processor TMS320C30 is applied.

For workpiece recognition, the image processing unit of an assembly robot must be fast and reliable. To improve an already existing system, we implemented an ASIC for real-time image correlation from a standard video image based on an investigation of several algorithms for 2D grayscale image correlation. The architecture contains four parallel multipliers and uses a three- level pipelining schema. The ASIC, which is realized in a standard and macrocell technique, achieves a sustained performance of 16 X 10⁶ MAC/s. A speed-up factor of 12 compared to a DSP56116 implementation at the same clock speed is obtained.

A vision-based system has been developed for industrial lace cutting. On-line pattern recognition is performed to detect the cutting path (river). The river is vectorized and passed to a numerical control system that drives a cutter. Since lace is subject to stretch as it is passed through the machine, vision is used to detect changes and provide the necessary cutter path compensation. Detection is carried out in real time and changes are fed back to the control subsystem. Various ways of change detection are investigated. The most effective method is found to be that of frame comparison using prior knowledge. Prior knowledge is used to enable rapid isolation of the river and detection of changes in the river pattern. Several experiments have been carried out using lace patterns of varying complexity. The experimental results are presented. A working prototype is described. The effectiveness of the optimization techniques employed to meet the real-time constraints is evaluated.

A technique for the quality assessment of fine-pitch electronic components is described. A system for the capture of Fraunhofer diffraction patterns reflected from the component is presented. The processing of the image to enable classification by a neural network is discussed. Simulation results are presented, showing the feasibility of the technique.