We describe an industrial inspection system with multiple data sources for object location and inspection. Within the context of this system, we present an approach for extraction of reliable and full object features consisting of both surfaces and space curves from registered range and intensity data. For matching, a homogeneous representation is adopted, which is applicable to a wide classes of object features. However, initial results are demonstrated on line and plane data only.
This paper describes the application of machine vision tech n iques to the problem of surface inspection of web products, in particular high-grade painted steel strip. A different approach has been adopted to the system architecture from that found in most contemporary systems. By u tilising tradi tional machine vision techniques embodied in a combination of off-the-shelf and custom vision processing technology, a highly modular system architecture has been developed which may be applied across .a broad spectrum of inspection tasks. Each inspection module within the system consists of a standard solid-state camera, commercial digitiser and frarne store and one or more custorn processing boards. These custom boards are configurable spatial filters which are inexpen sive to replicate, allowing multiple boards optirnised for subsets of the defect space to be used. This allows the processing power required to detect defects of interest to be tailored to the application. Similarly the number of complete detection modules can be controlled in a modular fashion to meet a pplication requirements. A real-time processor coordinates the detection rnodu les ancl handles defect parameterisation and image transfer to a host computer which classifies defects and provides a user i nterface for an operator. A fully functional two-module systern has been in operation on a production steel st.rip paint.line since mid 1991.
Machine vision has been characterized by a lack of standards, which has resulted in the necessity for expensive customization of systems. Our objective is to investigate the development of a vision system which utilizes or establishes standards. To this end we chose to use laser triangulation, to incorporate television broadcast standards, and to implement the software in an object-oriented class library encapsulating the essential features of the major components of a vision system. A 3D color digitization system was conceived and implemented, based on existing standards, with the flexibility and extensibility provided by object-oriented software design. The system generated data in a standard 3D file format, and was used to digitize and create rendered images of a building. Although the construction of classes needed careful planning, once created they greatly facilitated system modification. Existing standards are suitable for use in a 3D vision system, but there are several limitations which are considered.
We have developed an automated inspection system that features perpendicular and variable- intensity lighting for image contrast enhancement and improved sensing accuracy, a high- resolution camera with reflection-adaptive binarization for improved image processing, and an adaptive inspection algorithm that modifies its defect definition criteria according to target position quickly, accurately, and reliably inspects highly dense arrays of perpendicular I/O pins soldered onto a ceramic printed wiring board (PWB). The system's Mega-Scope, a high- resolution, eight-bit gray-scale CCD camera, images a 2048 X 2048-pixel area with a 10 micrometers resolution in 4 seconds, taking 60 I/O pin images at a time. The total time to inspect the position and solder fillet condition of more than eight thousand I/O pins is just 30 minutes.
Evaluation of cracks on the surface of the road is an essential work for public organizations that are in charge of maintenance of roads. Currently, men view the photographs of the road to evaluate the cracks of the road surface. The work requires much man-power: high skill and careful examination. So far, several efforts have been made to realize automatic evaluation system of the cracks. Many of them are however in the experimental stage because of the amount of image data. Our system is developed as a practical system for the actual use of the road crack evaluation. Our system has the following advantages: (1) Real-time image processing hardware. Road image is at most 1,000-feet long. It is difficult to handle with tall that amount of image on even today's workstations. A special real-time image processor reduces image data into 1/200 without losing an essential crack information. (2) Good human interface. We design the human interface carefully to achieve a truly easy-to-use system. (3) Efficiency. This system reduced the man power dramatically. This paper presents a general structure of the system, overview of the algorithms to detect and evaluate cracks, and discuss system integration including human interface issues.
A semi-automated technique for bore inspection of small diameter tubes is presented. The inspections are performed to insure that the bore surfaces are free of contaminants or defects. The image collection scheme uses a borescope which is stepped along the length of the tube. An image is acquired at each step and portions from each image are combined to yield a planar image. Color analysis classifies the oxidation levels in the bore of the fill tubes. The analysis is performed by taking the image's mean values of the red, green, and blue intensities and computing a figure of merit which is then used to estimate the relative amount of oxidation. This estimation scheme was shown to have a high level of correlation with the tube oxidation levels and the quality of the subsequent welds as determined by metallographic evaluation. Surface imperfections are detected by a series of digital filtering steps followed by a statistical analysis of the resulting binary image. The frequency parameter of the Poisson distribution for the total image and image segments are computed. A statistical significance test is performed by comparing the frequency parameter of each segment to the global statistics of the image. Fine longitudinal scratches were detected with this method.
Third Internal Visual Inspection in the IC assembly process deals with the wire bond quality inspection and is a very labor intensive stage. This paper describes a simple but comprehensive prototype model developed to automate the inspection of wedge and ball bonds. It is assumed that the die tilt is within specifications and the bond pad is automatically aligned. The defects associated with the bond quality are classified into four categories: size, shape, position and dimension. The bond is inspected sequentially for each category of defects and is rejected without further processing when any defect is detected. The procedure adopted in the prototype is as follows: A global thresholding technique automatically binarizes the intensity image of the bond. Bond limits are determined by a scanning procedure. Simple techniques such as pixel count, MER, Centroid and Median are used to verify the specifications related to size and shape. An intelligent scanning technique inspects the position related specifications in addition to identifying the wire and tail positions of the bond. The dimensions of the bond are determined by using projection information. 100% success is achieved with the experiments conducted using 100 sample images.
The problem of automated defect classification has been recognized as one of the biggest challenges to successful integration of automated inspection into wafer manufacturing process. The high degree of customization required for each application adds another dimension to the inherent difficulties. For example, the input to an automatic classifier of printed circuit boards defects will be completely different from the input to a semiconductor defect classifier. Also, the heuristics used for classification will differ greatly. Furthermore, even in similar applications, different manufacturers will have different notions of how defects ought to be classified. In this paper we describe a system which attempts to automate the defect classification process, offering a high degree of adaptivity, customization and automation, with special emphasis on minimizing the need for input from a skilled user. Using Orbot's wafer inspection system, we concentrated on classification of defects on patterned wafers.
The automatic, real-time visual acquisition and inspection of VLSI boards requires the use of machine vision and artificial intelligence methodologies in a new "frame" for the achievement of better results regarding efficiency, quality and automated service. Thus, the development of a knowledge base for automated inspection and acquisition of VLSI boards becomes an important issue with today's needs from the electronic marketing. Inthis paper we describe the visual inspection of damaged VLSI boards and discuss the specifications for the development of a knowledge base associated with the visual inspection.
While initial detection of defects is the most critical function of inspection, automatic classification of detected defects is becoming increasingly desirable. The key to better process control is reliable process measurement. The classification of defects provides valuable process diagnosis information. The hope is that machines can perform this task more reliably than humans. However, there are many problems in automating defect classification, and many of these are related to the central problems in artificial intelligence, such as knowledge representation, inferencing, and dealing with uncertainty. In this paper we pay special attention to the issues arising in the Automatic Defect Classification (ADC) of integrated circuits. We first discuss technical and system requirements, followed by an outline of the technical challenges to be overcome to develop flexible and powerful ACD tools which can be quickly customized on a user level for diverse applications.
This paper describes the development environment for the design, implementation, and testing of an expert system for web inspection that uses adaptive digital image processing and pattern recognition to classify defects in web materials. The environment consists of four sub-systems: detection, characterization, feature analysis, and classification. Each sub-system consists of a number of modules targeted at performing specific functions. The paper discusses an example of defect classification and describes the roles of characterization, feature analysis, and classification sub-systems, including specific algorithms.
We will discuss about some simple features for automatic inspection of surfaces whose defect patterns are aggregations of irregular shapes. The description or classification of these defects is not an easy task. Two types of feature sets are studied--a set of features based on connected component labeling, and a set of local measurements that can be computed easily. As the first set, several region properties that can be computed from labeled binary images have been tested. Each of these features is weighted according to its variance and dependencies with respect to other. A classification method based on the minimum difference between the trained data and the distribution computed from an image of an unknown class have been used to test the feature set. As the second set, we have defined about 10 features that can be computed without labeling the binary image. By using classification method based on the distribution of feature values along with weighting factors, we have obtained a high rate of correct classification for 20 classes of complex natural images.
A self-reference technique is developed for detecting the location of defects in repeated pattern wafers and masks. The application area of the proposed method includes inspection of memory chips, shift registers, switch capacitors, CCD arrays, and liquid crystal displays (LCD). Using high resolution spectral estimation algorithms, the proposed technique first extracts the period and structure of repeated patterns from the image to sub-pixel resolution, and then produces a defect-free reference image for making comparison with the actual image. Since the technique acquires all its needed information from a single image, there is no need for a database image, a scaling procedure, or any a-priori knowledge about the repetition period of the patterns. Results of applying the proposed technique to real images from microlithography are presented.
The use of ultrasonic imaging to analyze defects and characterize materials is critical in the development of non-destructive testing and non-destructive evaluation (NDT/NDE) tools for manufacturing. To develop better quality control and reliability in the manufacturing environment advanced image processing techniques are useful. For example, through the use of texture filtering on ultrasound images, we have been able to filter characteristic textures from highly-textured C-scan images of materials. The materials have highly regular characteristic textures which are of the same resolution and dynamic range as other important features within the image. By applying texture filters and adaptively modifying their filter response, we have examined a family of filters for removing these textures.
PC-based inspection systems for wide web materials have been unable to effectively image fine defects as they are detected. The amount of data produced by highly parallel video inspection cameras can exceed 400 MBytes/sec. The system described in this paper is capable of analyzing and displaying a detected image within seconds of the even using a single frame grabber and a 386 computer. The system can operate at processing speeds of greater than 400 MBytes/sec since it makes use of a novel post processing algorithm within the camera itself. The video cameras are based on TDI (Time Delay and Integration) technology to provide high grey scale resolution at high data rates and low light levels. The system has an adjustable resolution ranging from 2000 to 24,000 pixels per line scanned. The scanning rate is adjustable to a maximum of 20,000 line scans per second.
Two dimensional Area Parameters are providing adaptable and robust in machine vision applications when implemented with hardware algorithms to insure performance. Let's review the basics of Area Parameter implementation, then briefly view some applications: steel web inspection, road and pavement inspection, traffic analysis, surface analysis for fine coatings, Non Destructive Testing, Security, Production Laser Control, Currency Inspection, Automated Microscopy, and Agriculture and Picking.
In this paper we present an overview of the state of the art in SXM with emphasis on image processing techniques for SXM. We outline the principle of operation of different scanning probe microscopes. Issues related to sensor technology are discussed. Commercially available scanning probe microscopes are listed and their features summarized. We review in detail the image processing work that has been done to date in relation to SXM and raise relevant issues. Existing and potential applications of SXM are discussed. Finally, we point out directions for future research in image processing related to SXM.
We describe an experiment in which the etch depth of a diffraction grating is measured. A simulated experiment is used to develop and calibrate the measurement technique. A scatterometer was used to measure the diffraction patterns of a set of 5 wafers at 14 die locations. The estimator already developed is then used to find the etch depths at the 70 measured locations. Finally, a scanning force microscope is used as a reference method to validate the scatterometer measurements.
A technique using diffraction grating test patterns has been used to monitor in situ the pre and post etch linewidths on a polysilicon etch chamber. The technique is capable of linewidth measurements to 0.25 microns with pitches as small as 0.7 microns. A comparison of in situ polysilicon linewidth measurements with off-line SEM measurements shows measurement differences of less than 10% indicating a measurement precision on that order. The repeatability of the diffraction technique is shown to be approximately 0.01 micron in comparison to the typical SEM repeatability of 0.025 micron. The implementation of this technique on a production etch chamber required the design of specialized optics and image processing systems. The optics system allows the monitoring of one and a half dimensions of the diffraction pattern. It consists of a CCD camera and some reflective optics for the focusing of the diffraction patterns. The image processing system uses a commercially available frame grabber and several image processing algorithms to record these patterns and extract the linewidth information. Algorithms unique to this application include an image indexing scheme used to store the diffraction images, a blob classification scheme based on a rigorous examination of the three dimensional vector field problem, and a non-linear iterative modeling algorithm used to fit the data to Fourier theory predictions. The resulting system is capable of linewidth measurements on each wafer with an overall reduction in product cycle time relative to the previously used SEM and pilot monitoring schemes.
A microscope coherent optical processor (M-COP) has been specifically applied to the measurement of registration error in multi-layer integrated circuit wafers. This novel technique has resulted in a simple and elegant method for the inspection of both part- and fully-processed semiconductor wafers. The conventional methods of processed wafer inspection comprise of microscopical examination by skilled operators, extensive electrical parameter testing, and numerical image-processing. To apply any one of these techniques to a single chip on a wafer is extremely time consuming, and the number of samples that can be inspected on a production line is severely limited. However, real-time 100% inspection of wafers is one of the benefits of the inherent high resolution and immense speed of the M-COP. Further benefits are the early detection and location of faults, 100% quality assurance, and continuous condition monitoring.
We will describe in this paper work that is geared towards the development of a tool to assist in the visual tasks of grading and inspecting poultry. Extensions to similar activity in other food processing arenas will be discussed. We specify aids since we believe that these systems will not be able to fully conduct the range of functions required but would be able to provide capable assistance if they function in a screening capacity.
The goal of our research is to develop an automated system for the visual inspection of wood surfaces. Our present approach to inspection is based on the computing of color and texture features for non-overlapping image windows and classifying each window into one of a set of prototype defect classes on the basis of training statistics. The choice of features and final classification strategy nevertheless requires an analysis of the color properties of defects and clear wood. The spectra reflectance characteristics of objects in the visible spectrum were used in the color analysis, these being invariant properties of objects concerned. The analysis was performed for pine wood, employing measurements of the spectral reflectance characteristics of clear wood and visible defects. This enabled a few types of clear wood to be classified in terms of the degree of color homogeneity, defects recognizable against the background of the clear wood and defects differentiable by color. The results indicate that color provides very valuable information for the discrimination of wood defects. Such spectral reflectance characteristics of defects can also be used to formulate the requirements for a machine vision system.
A computerized inspection system is described, which automatically characterizes image defects, evaluates image quality statistics, and compares the result with previously stored specifications. The algorithm and implementation are focused on agreement with empirical observations, reliability, and ease of use.