The identity of Machine Vision as an academic and practical subject of study is asserted. In particular, the distinction between Machine Vision on the one hand and Computer Vision, Digital Image Processing, Pattern Recognition and Artificial Intelligence on the other is emphasized. The article demonstrates through four cases studies that the active involvement of a person who is sensitive to the broad aspects of vision system design can avoid disaster and can often achieve a successful machine that would not otherwise have been possible. This article is a transcript of the key- note address presented at the conference. Since the proceedings are prepared and printed before the conference, it is not possible to include a record of the response to this paper made by the delegates during the round-table discussion. It is hoped to collate and disseminate these via the World Wide Web after the event. (A link will be provided at http://bruce.cs.cf.ac.uk/bruce/index.html.).

A vision system to gauge two types of automotive parts has been developed. One of the part types is a power steering connector in which the depth and width of the groove, and the distance between the start of the groove and the end of the power steering line are gauged. For the second type of part, a crimped connector attached to a brake hose, measurements of interest are the two diameters of the crimp and the bell length where the hose is inserted into the connector. A standard video camera is used to acquire the image of a back-illuminated image of the part which is digitized and captured with a frame grabber. The basic hardware to accomplish the gauging tasks consists of a standard video camera, light source, frame grabber and industrial personal computer. In order to minimize hardware costs, a standard 50 mm C-mount camera lens and extension tube was used with the video camera. Consideration had been made to use more expensive telecentric optics so that parts placement would not cause a change in magnification with a resulting loss of accuracy. With the 50 mm lens, however, magnification effects were lessened due to the greater standoff distance between camera and part. For image acquisition, a low-cost PCI-bus frame grabber-card was chosen. With this type of card, high-speed video capture is possible due to the very wide bandwidth of the PCI bus. Combined with a Pentium-based PC, rapid image acquisition and analysis can be performed so that every part can be gauged at full production rates. Since the gauging rate exceeds the production rate by a significant factor, a single computer and frame grabber with camera multiplexer can process data in real time from up to four measurement stations simultaneously.

This paper describes the benchmarking of image processing algorithms using high-performance workstations and personal desktop computers. For the various platforms evaluated which included machines from Sun, SGI, Apple, and Gateway, compiler options were varied to obtain the fastest execution times. Algorithms evaluated included typical image processing operations such as derivatives, logical operations, morphology, subtraction, median filter, and the new SKIPSM approach. Data were collected using the different platforms and are presented here in tabular form. The results indicate that the latest generation of personal computers have processing capabilities that are similar to UNIX-based work stations.

3D data present pertinent information about geometrical features of an object. It has been a classical approach that image data acquired by range sensors are processed as traditional 2.5D images. Range data have rich information that needs some special treatment in order to fully understand and utilize them. In this report, two case studies are presented to investigate the 3D aspects of range data and applied them to solve practical problems in manufacturing environment. The first case is the classical `pick and place' problem where the range data were taken on the holding rack of car doors and there is a need to identify multiple holding points accurately, with the rack almost taking entire field of view and traditional image processing approach invalid. The second case is the range data correction and sub-pixel patch with large field of view. The applications are in automated quality assessment. Two approaches are from different vision solutions with one objective that is to process range image data with 3D representation instead of that of traditional 2.5D.

During the past year, several new computer camera methods (hardware and software) have been developed which have applications in machine vision. These are described below, along with some test results. The improvements are generally in the direction of higher speed and greater parallelism. A PCI interface card has been designed which is adaptable to multiple CCD types, both color and monochrome. A newly designed A/D converter allows for a choice of 8 or 10-bit conversion resolution and a choice of two different analog inputs. Thus, by using four of these converters feeding the 32-bit PCI data bus, up to 8 camera heads can be used with a single PCI card, and four camera heads can be operated in parallel. The card has been designed so that any of 8 different CCD types can be used with it (6 monochrome and 2 color CCDs) ranging in resolution from 192 by 165 pixels up to 1134 by 972 pixels. In the area of software, a method has been developed to better utilize the decision-making capability of the computer along with the sub-array scan capabilities of many CCDs. Specifically, it is shown below how to achieve a dual scan mode camera system wherein one scan mode is a low density, high speed scan of a complete image area, and a higher density sub-array scan is used in those areas where changes have been observed. The name given to this technique is adaptive sub-array scanning.

High accuracy digital image based metrology must rely on an integrated model of image generation that is able to consider simultaneously the geometry of the camera vs. object positioning, and the conversion of the optical image on the sensor into an electronic digital format. In applications of automated visual inspection involving the analysis of approximately plane objects these models are generally simplified in order to facilitate the process of camera calibration. In this context, the lack of rigor in the determination of the intrinsic parameters in such models is particularly relevant. Aiming at the high accuracy metrology of contours of objects lying on an analysis plane, and involving sub-pixel measurements, this paper presents a three-stage camera model that includes an extrinsic component of perspective distortion and the intrinsic components of radial lens distortion and sensor misalignment. The later two factors are crucial in applications of machine vision that rely on the use of low cost optical components. A polynomial model for the negative radial lens distortion of wide field of view CCTV lenses is also established.

This paper describes a new technique for stereoscopic 3D position measurement. By defining stereo cameras as a system for image-to-world mapping, the mapping function is determined. The direct representation of 3D coordinates of a world point with corresponding stereo image coordinates is derived using the pin-hole model. One camera frame is related to the other before being related to the world frame so that the stereo itself rather than each camera can be directly related with the 3D world. The equations obtained are simple, straightforward, and closed-form. However, since the nonlinearity of actual imaging system is not considered, high accuracy is difficult to be expected when the equation is employed for 3D measurements. For tackling this problem a multilayer feedforward neural network trained by back- propagation algorithm is used. The network played the role of fine correction satisfactorily using its function learning capability after rough mapping by the linear equations in our experiment.

Circuit and theory of a new type of the polarization device is described. Introduction of an additional return beam allows to stabilize optical phase and index of modulation and, in addition, simultaneously to determine gyrotropic and anisotropic characteristics of researched object.

A twin-orthogonal-fanbeam x-ray system has been built as part of a six-partner project funded by the Commission of the European Union. The images created by this system represent plan and side views of the object to be inspected. Using such a system, it is possible to locate a point-like feature that creates a significant shadow in both beams, in a 3D space. However, the real value of such a system lies in the fact that it is often possible to see a foreign body, such as a small piece of loose glass, within a jar using one beam, when the same contaminant is invisible to the other beam. Such a situation typically arises when the foreign body is obscured by the x-ray shadow of the neck-shoulder region of a jar. The x-ray system built by our colleagues in this consortium is being used to examine, simultaneously, six jars of semi-fluid savory sauce, held together by shrink-wrapping on a cardboard tray. The inspection algorithm consists of fitting multi-part models of the image intensity function to both the plan and side-view images. Once a model has been fitted, it is possible to use image comparison, in order to highlight any foreign bodies. The pre-processed plan and side-view images are analyzed and correlated together, so that in many cases, a foreign body whose view is obscured in one image can be detected in the other.

This paper describes the implementation of INFIBRA, a machine vision system for the inspection of acrylic fiber production lines. The system was developed by INETI under a contract from Fisipe, Fibras Sinteticas de Portugal, S.A. At Fisipe there are ten production lines in continuous operation, each approximately 40 m in length. A team of operators used to perform periodic manual visual inspection of each line in conditions of high ambient temperature and humidity. It is not surprising that failures in the manual inspection process occurred with some frequency, with consequences that ranged from reduced fiber quality to production stoppages. The INFIBRA system architecture is a specialization of a generic, modular machine vision architecture based on a network of Personal Computers (PCs), each equipped with a low cost frame grabber. Each production line has a dedicated PC that performs automatic inspection, using specially designed metrology algorithms, via four video cameras located at key positions on the line. The cameras are mounted inside custom-built, hermetically sealed water-cooled housings to protect them from the unfriendly environment. The ten PCs, one for each production line, communicate with a central PC via a standard Ethernet connection. The operator controls all aspects of the inspection process, from configuration through to handling alarms, via a simple graphical interface on the central PC. At any time the operator can also view on the central PC's screen the live image from any one of the 40 cameras employed by the system.

Morphological granulometric moments have proven useful for quantification and classification of image texture. This paper considers their use as measures of surface roughness. The analysis is based on simulations in the framework of a modified Boolean random function model for surfaces. Four granulometric features are considered: the pattern-spectrum mean and pattern-spectrum variance for both opening and closing granulometries generated by flat structuring elements. The expectations of these granulometric moments are compared with the expectations of the classical average and root-mean-square roughness across a range of Boolean models.

For many years, manufacturers have been both managing and assuring the quality of their products. Among the many concepts used at present, artificial vision is chosen to control conformity during the production process. The real time control of 3D reflective objects is our concern in this research work. In the recent past, work has been done showing the capability of a vision system for detecting and measuring defects on non-moving 3D reflective manufactured products. For this new industrial application, defects we are trying to detect are dusts included under the metallic layer of products. The defect detection is still based on the use of a specific lighting system composed of light stripes. By examining the reflection of light from object, we are able to detect defects between two consecutive white stripes. The optimal detection has been obtained by adjusting the characteristics of the lighting system while taking into account the 3D geometry of the defects to be detected. The machine vision system performs the control of one cylindrical product within one second.

3D coordinates acquisition and 3D model generation for existing parts or prototypes are the critical techniques in reverse engineering. This paper presents an integrated intelligent inspection system of stereo vision and coordinate measurement machine which is fast, flexible and accurate for reverse engineering. It also emphatically discusses the principle, structure and key technique of the system.

To construct highly precise space structures, such as antennas, it is essential to be able to collimate them with high precision by remote operation. Surveying techniques which are commonly used for collimating ground-based antennas cannot be applied to space systems, since they require relatively sensitive and complex instruments. In this paper, we propose a collimation method that is applied to mark-patterns mounted on an antenna dish for detecting very slight displacements. By calculating a cross- correlation function between the target and reference mark- patterns, and by interpolating this calculated function, we can measure the displacement of the target mark-pattern in sub-pixel precision. We developed a test-bed for the measuring system and evaluated several mark-patterns suitable for our image processing technique. A mark-pattern with which enabled to detect displacement within an RMS error of 1/100 pixels was found. Several tests conducted using this chosen pattern verified the robustness of the method to different light conditions and alignment errors. This collimating method is designed for application to an assembling-type antenna which is being developed by the Communications Research Laboratory.

This paper discusses user-interface and data structures for a triggered vision system, the Engine-VideoScope 513D. Usually, vision-systems for sophisticated evaluation purposes provide separate software-products for system configuration, online-inspection, digital image processing, and presentation. Individual products for various tasks originate only from the developers--but not from the customers-point of view. The user has to work with powerful tools although only a few features might be sufficient for his work. Instead of concentrating on his problem, he repeatedly has to operate with applications from various vendors, interfaces, data management and documentation. Our approach is not just to provide an application for image acquisition. We referred to the users needs for a delimited field of applications, considering both, hardware technology and typical evaluation steps. We will show, that a properly designed user-interface can drastically improve the working- efficiency and the level of acceptance from the users. The Engine VideoScope was originally designed to inspect and interpret processes in the field of powertrain engineering. However, providing an open software architecture and data- structures could be suitable for prototyping machine vision systems or even the base for products in various fields.

In the field of marine biology, determining the presence and quantities of different types of fish is traditionally done by dragging nets across the bottom, and counting that which is found in the nets. This method, although accurate, kills the collected fish, damages the habitat, and consumes large quantities of time. This paper presents an alternative. A machine vision system is capable of counting and measuring fish in an ocean environment. Illumination presents a unique problem in this environment. Object orientation and measurement are related and resolved issues. An adaptive thresholding technique is required to appropriately segment the fish from the background in the images. Mode detection, and histogram analysis are useful tools in determining these localized thresholds. This system, created in conjunction with the Rutgers Institute for Marine and Coastal Science, effectively counts and measures fish in an estuarine environment.

A real-time face tracker is presented in this paper. The system has achieved 15 frames/second tracking using a Pentium 200 PC with a Datacube MaxPCI image processing board and a Panasonic RGB color camera. It tracks human faces in the camera's field of view while people move freely. A stochastic model to characterize the skin color distribution of human skin is used to segment the face and other skin areas from the background. Median filtering is then used to clean up the background noise. Geometric constraints are applied to the segmented image to extract the face from the background. To reduce computation and achieve real-time tracking, 1D projections (horizontal and vertical) of the image are analyzed instead of the 2D image. Run-length- encoding and frequency domain analysis algorithms are used to separate faces from other skin-like blobs. The system is robust to illumination intensity variations and different skin colors. It can be applied to many human-computer interaction applications such as sound locating, lip- reading, gaze tracking and face recognition.

The work offers the methods for invariant representation of images against a variety of distorting factors including 2D and 3D rotation, changes in brightness, contrast and scale. The problems of preliminary image processing based on the method of generalized Q-transformation are being solved. The calculating algorithms based on the methodology of dichotomous balance of the images being prepared have been used for the classification of human facial images. It also deals with the procedure of recursive contour preparation consisting of step-by-step preparation of differences among the pixels of grey-scale image and formation of positive, negative and zero preparations. Thus at the first step the contour preparation is effected for the first differences, at the second step, for the second differences, and so on, with a step-by-step definition of the criterial function of distribution of binarized preparations. So it is possible to identify objects in different lighting conditions which simplifies the implementation of similar approaches. This relative simplicity of this method extends the range of its possible application for recognition purposes and for its implementation in the parallel-hierarchical network in particular.

In this paper, a new on-line measurement and accuracy analysis method for part configuration and surface is presented by combining computer vision and neural networks. Different from conventional contact measurement, it is non- contact measurement method, and it can operate on-line. In this method, the 3D configuration and surface of part are reconstructed from stereo image pair taken by computer vision system. The architecture for parallel implementation of part measurement system is developed using neural networks. Several relevant approaches including system calibration, stereo matching, and 3D reconstruction are constructed using neural networks. Instead of conventional system calibration method that needs complicated iteration calculation process, the new system calibration approach is presented using BP neural network. The 3D coordinates of part surface are obtained from 2D points on images by several BP neural networks. Based on the above architecture and the approaches, the part measurement and accuracy analysis system for intelligent manufacturing is developed by making full use of the advantages of neural networks. The experiments and application research for this system is also presented in this paper. It is proved through the actual application that the method presented in this paper can meet the needs of on-line measurement for parts in intelligent manufacturing. it has important value especially for on-line measurement of parts that have complicated surface.

In this work, we present an inspection method for power and ground (P&G) layers of printed circuit boards (PCB) also called utility layers. Design considerations for the P&G layers are different than those of signal layers. Current PCB inspection approaches cannot be applied to these layers. P&G layers act as internal ground, neutral or power sources. P&G layers are predominantly copper with occasional pad areas (without copper) called clearance. Defect definition is based on the spacing between the holes that will be drilled in clearances and the surrounding copper. Overlap of pads of different sizes and shapes are allowed. This results in complex, hard to inspect clearances. Our inspection is based on identification of shape, size and position of the individual pads that contribute to an overlapping clearance and then inspection of each pad based on design rules and tolerances. Main steps of our algorithm are as follows: (1) extraction and preprocessing of clearance contours; (2) decomposition of contours into segments: corner detection and matching lines or circular arcs between two corners; (3) determination of the pads from partial contour information obtained in step (2), and (4) design rules checking for each detected pad.

A visual information directed microphone array system is presented in this paper. This system uses a real-time mouth tracking system to direct a beam-former focusing on the mouth. The microphone array system is implemented on a PC with a Signalogic 8-channel DSP board and reports a better signal-to-noise ratio sound capturing in a high noise environment.

This paper describes how under-resolved images of bar codes may be read by suitable processing. A 2D image of a bar code with insufficient resolution to be able to resolve the individual bars is processed to give a high-resolution image. For this to work, the bar code (or camera) must be slightly rotated to give a fraction of a pixel offset between rows. Since the bars are straight, the offset relative to the first complete row of the bar code increases linearly with vertical position in the image. This offset between rows results in a shift in phase that is proportional to both offset and frequency. A phase image is formed by Fourier transforming each row in the image, and retaining the phase. By subtracting the first row from subsequent rows of the phase image, a surface is fitted to give the offset between rows. A high-resolution image is then formed by interleaving the pixel values from rows where the offset is nearest to the new pixel spacing. This image appears blurred because of the area sampling caused by the sensor, combined with the low pass response of the camera electronics. By modeling the image capture system, the point spread function may be estimated and then removed by using inverse filtering in the frequency domain. The offset between the rows is then removed by using a linear phase filter. This allows the rows within the resultant image to be averaged to reduce noise.

This paper describes the design of a PC-based real-time machine vision system for detecting and classifying small marine organisms like fish eggs and planktons in flowing water. The system is called the Real-time FLow Imaging and Classification System, or ReFLICS for short, and it will automate the task of visually counting and classifying fish egg samples which is currently performed by trained humans. ReFLICS uses a line-scan image sensor to eliminate double counting and boundary effects. Using a combination of flowmeter and image-based flow error correction algorithm, ReFLICS's line-scan camera can work with changing flow. Design of the complete system from the camera and illumination housing to the machine vision software allows ReFLICS to work in the harsh environments of a ship at sea. Using an industry-standard multi-processor PC with PCI card pipeline image processor and running Microsoft Windows NT, ReFLICS can achieve the high performance required meanwhile maintaining relative low equipment, development, and maintenance costs. This paper provides the ReFLICS system design and presents initial results of the system.

With increased demand for reliable and automated semiconductor wafer inspection, machine vision techniques are much needed for defect extraction and identification and pre- and post-processing operations. The wavelet-based approach is emphasized in this paper for its capability to meet different needs in wafer inspection. The algorithms involved are illustrated by using a number of wafer inspection images.

One step in the manufacture of circular fluorescent lamps is the mercury (Hg) injection process in which a small amount of mercury approximately 20 mg is injected into the fluorescent tube. An on-line detection of mercury is required to ensure that the amount of mercury residual inside the tube is within the acceptable tolerance. This critical operation is to reduce manufacturing overhead by controlling cost and reducing waste of materials. In view of this, an attempt has been made to design and develop an on- line mercury detection system with important benefits to the manufacturers, such as increased throughput, accuracy, reliability and consistency. This paper presents the configuration and development works of the on-line circular fluorescent lamp inspection system developed by Industrial Project Group--Machine Vision Center of Nanyang Polytechnic. The inspection system performs on-line detection of mercury particles (Hg) inside the circular fluorescent lamp. Taking the orientation and translation offsets of the lamp into consideration, it detects the presence/absence as well as the size of the injected mercury. The system has been successfully operating 15 hours per day and inspecting more than 22 thousands lamps in the production plant.

A research group at the University of Stuttgart has set up an experimental measurement robot for industrial close range inspection. During a test run, the feasibility of a multi- sensor/actor system has been shown. The system uses optical sensors to perform different tasks including object recognition, localization and gauging. It is a step towards systems which are able to inspect and gauge several parts from a set of parts stored in a 3D model database. This paper describes the results which have been obtained so far and were demonstrated during a test run. It then focuses on our latest developments concerning 3D data acquisition, registration, segmentation, model generation from CAD data and object recognition.

X-ray laminography and DT (digital tomosynthesis) are promising technologies to form a cross-section image of 3D objects and can be a good solution for inspection interior defects of industrial products. It has been known that digital tomosynthesis method has several advantages over laminography method in that it can overcome the problems such as blurring effect or artifact. The DT system consists of a scanning x-ray tube, an image intensifier as an x-ray image detector, and a CCD camera. To acquire an x-ray image of an arbitrary plane of objects, a set of images (8 images or more) should be synthesized by averaging or minimally calculating point by point. The images, however are distorted according to the configurations of the image intensifier and the x-ray source position. To get a clear and accurate synthesized image, the corresponding points in the distorted images should be accurately determined, and therefore, precise calibration of the DT system is needed to map the corresponding points correctly. In this work, a series of calibration methods for the DT system are presented including the correction of the center offset between the x-ray and the image intensifer, the x-ray steering calibration, and the correction of the distortion of the image. The calibration models are implemented to the DT system and the experiment results are presented and discussed in detail.

We describe a PC-based machine vision system, developed for fast and precise measurement of cooking plate dimensions. To achieve the required accuracy two cameras with telecentric lenses, sub-pixel techniques and semi-automatic system calibration was used. The system has been integrated into the factory production line and is operating in real-time. The results obtained so far confirm that the system works reliably and that it meets the required accuracy.

Most of the published work about SKIPSM (Separated-Kernel Image Processing using finite-State Machines) has concentrated on large-neighborhood operations (e.g., binary morphology, Gaussian blur), because the speed improvements are the most dramatic in such cases. However, there are many frequently, used 3 X 3 operations that could also benefit from speed improvements that arise from separability and the use of finite-state machines. This paper shows SKIPSM implementations for an extensive list of 3 X 3 operations, including various edge detectors, connectivity detectors, direction of brightest neighbor, largest gradient, and direction of largest gradient. A generic implementation applicable to all 3 X 3 binary operations, whether separable of non-separate, is also given.

Median filters and ranked filters of ranks other than median have often been proposed or used to remove image noise as well as for other reasons. These are nonlinear operations, and often have relative long execution times, making them unsatisfactory for many speed-critical industrial applications. This paper builds on the earlier work of Mahmoodi and Waltz to provide efficient implementations of 3 X 3 ranked filters of ranks 1 (minimum), 2, 3, 4, 5 (median), 6, 7, 8, and 9 (maximum). These implementations are based on a partial realization of the SKIPSM (Separated- Kernel Image Processing using finite-State Machines) paradigm. A full SKIPSM realization is not possible because, except for the filters of ranks 1 and 9, these operations are not separable. This paper shows that, in spite of this lack of separability, the finite-state machine aspect of SKIPSM can be used to advantage. The emphasis is on software implementations, but implementation is pipelined hardware have also been demonstrated. In addition, a fast `full- SKIPSM' implementation of a slightly different ranked filter, sometimes called the `separable median' filter, is presented. This filter guarantees that the output pixels are of rank 4, 5, or 6. For typical noise-reduction applications, it is difficult to find a convincing argument that this filter is inferior in any meaningful way to the true median filter.

The SKIPSM (Separated-Kernel Image Processing using finite- State Machines) paradigm has been extended with excellent results to grey-scale morphology with arbitrary flat structuring elements. But casual users can not be expected to master the techniques for creating SKIPSM implementations for user-specified SEs, thus limiting the usefulness of the technique. This paper addresses that limitation by providing a completely automated procedure for generating the SKIPSM implementation for a wide range of grey-scale image processing operations in addition to grey-scale morphology, given only the definition of the neighborhood over which the computations are to be made. These neighborhoods need not be square or rectangular, but can be made up of arbitrary collections of contiguous or non-contiguous pixels. Examples of the operations that can be performed include grey-scale dilation and erosion with flat structuring elements, area sum or arithmetic mean, geometric mean, etc. In effect, this paper provides the basic structure for a computer program to write efficient computer code in a target language such as C. This code-generating program could be written in almost any computer language, but because it involves both list processing and some backtracking. Prolog would be an excellent choice. This technique can be extended to three- and higher-dimensional grey-scale morphology without great difficulty.

The SKIPSM (Separated-Kernel Image Processing using finite- State Machines) has recently been extended with excellent results to various grey-scale operations (morphology, Gaussian Blur, ranked filters, etc.) as well as 3D binary operations. Another interesting direction, initiated in this paper, is the extension to color spaces. After a discussion of some possible goals of such image processing, this paper presents implementation techniques and examples. Because of the extra costs involved with the printing of colored images in journals, the example images are presented here in grey- scale only. Color images will be provided via e-mail on request.

Interactive image processing is a proven technique for analyzing industrial vision applications and building prototype systems. Several of the previous implementations have used dedicated hardware to perform the image processing, with a top layer of software providing a convenient user interface. More recently, self-contained software packages have been devised and these run on a standard computer. The advent of the Java programming language has made it possible to write platform-independent software, operating over the Internet, or a company-wide Intranet. Thus, there arises the possibility of designing at least some shop-floor inspection/control systems, without the vision engineer ever entering the factories where they will be used. It successful, this project will have a major impact on the productivity of vision systems designers.

There exists a serious bottle-neck in the process of designing Machine Vision Systems. This is so severe that the long-claimed flexibility of this technology will never be realized, unless there is a significant increase in the capacity of present-day vision system design teams. One possible way to improve matters is to provide appropriate design tools that will amplify the efforts of engineers who lack the necessary educational back-ground. This article describes a major extension to an existing program, called the Lighting Advisor, which is able to search a pictorial database, looking for key-words chosen by the user. The revised program bases its advice on a description of the object to be inspected and the working environment. The objective of this research is to reduce the skill level needed to operate the program, so that an industrial engineer, with little or no special training in Machine Vision, can receive appropriate and relevant advice, relating to a range of tasks in the design of industrial vision systems.

The work has set out to investigate the application of data compression on real-time images used in shape measurement and machine vision applications. The quality of monochromatic images produced from compression based on the lossless schemes and DCT transform were analyzed for their degradation level. The lossy based DCT method appeared to provide the higher compression ratio of 6:1 required. Special concern was focused into the fringe pattern analysis area, whereby how the degraded compressed fringe images could possibly effect the accuracy of its application output. The real-time image compression mechanism is anticipated for a seamless transmission to a personal computer, through a standard interface channel.

2D Gaussian blur operations are used in many image processing applications. The execution times of these operations can be rather long, especially where large kernels are involved. Proper use of two properties of Gaussian blurs can help to reduce these long execution times: (1) Large kernels can be decomposed into the sequential application of small kernels. (2) Gaussian blurs are separable into row and column operations. This paper makes use of both of these characteristics and adds a third one: (3) The row and column operations can be formulated as finite-state machines to produce highly efficient code and, for multi-step decompositions, eliminate writing to intermediate images. This paper shows the FSM formulation of the Gaussian blur for the general case and provides examples. Speed comparisons between various implementations are provided for some of the examples. The emphasis is on software implementations, but implementations in pipelined hardware are also discussed. Straightforward extensions of these concepts to 3- and higher-dimensional image processing are also presented. Implementation techniques for DOG (Difference-of-Gaussian filters) are also provided.

Alphanumerics and other characters can be decomposed into a minimal number of components, namely, line segments and, circular and elliptical arcs. The combination and relative location of these components (i.e. the character signature) uniquely determine the character identity. We are developing a pattern recognition engine, Software Engineering Engine (SEE), which computes the set of all line segments, circular and elliptical arcs that a given digital curve represents. From this obtained set, the original line segment or geometric arc that best fits the digital curve is extracted. Thus, the underlying shape of the digital curve can be determined with subpixel accuracy. SEE computes all this in linear time in the number of pixels in the digital curve. To further recognize characters, SEE will determine the linear, circular and elliptical components that comprise each character. SEE will then compare this character signature from the image with signatures in a character-signature database to secure the best fit. This approach has applications to the interpretation of engineering symbols and will be extended to interpret dimensionality associated with geometric objects indicated in an engineering sketch.

Color and texture have long been used as image features to segment and classify images. In most of the previous approaches, color and texture are used as two uncorrelated features, while in the real world, the spatial information and spectral information of an image are often tightly coupled. An feature extraction algorithm is studied in this paper, which represents colored texture in a unified way. With this approach, different spectral channels are correlated spatially to give an unified representation of both the color and texture information. In order to use this feature in image segmentation applications, properties of the feature are studied. A novel segmentation algorithm is proposed based on the study. Preliminary segmentation results are presented.

We have developed a high frame rate image display system to study attentional control and information capacity limitations for perception of static objects in a visual display. The system presents images at 114.4 frames/sec using the stereo mode of a video display monitor and the Datacube MV200 image processing system. The proposed experimental paradigm is an extension of previous work where numeric icons were displayed at `Stimulus Onset Asynchrony' (SOA) of as low as 16.7 msec/icon. With our high frame rate display system, we can achieve lower SOAs of 8.7 msec/icon and hence further examine the perceptual capabilities for short duration displays of static objects.

An algorithm is developed to determine the significant junction points of branching tubules during embryonic lung development. This algorithm involves a multidisciplinary approach to preprocessing the input data so that it is ready for such analysis. In addition to specific biochemical techniques employed (such as fluorescent staining), a number of image processing methods are utilized to remove noise and insignificant structures in the image. The final junction analysis is performed on the skeleton of the major branches using a variation of the medial axis transform. The significance of this research is both to enable searches in digital libraries of such biological data for branching during lung development as well as indicating to the biologist the important sites to consider for further investigation.

The non-destructive tests allow to establish the physical and structural conditions of a mechanical part, to verify its condition, the superficial wear and tear and then evaluate its `remaining' efficiency. The non-destructive tests are applied in all those fields of engineering in which the determination of the mechanical and structural characteristics of elements in use is requested, without making them undergo destructive or damaging tests. In the present work an application program has been developed which, examining the surface of mechanical parts under an optical microscope and a blaster video, is able to characterize the material and to recognize and identify the possible presence of a superficial crack. The program constitutes the first step towards the realization of an industrial prototype which, thanks to the utilization of a plan moved by step-by-step motors, allowing the scanning of the whole surface of a part and the recognition of the crack in an automatic way, that is without the presence of an operator, and its characterization, in case it is identified, through the determination of some geometric parameters useful to ascertain the structural integrity of the element under examination. For the realization of the program different techniques of image analysis have been applied and the use of an artificial neural network preset for the recognition of the crack has been necessary. The program has been realized in C language and it works in Linux system.

Color measurement is an important part of overall production quality control in textile, coating, plastics, food, paper and other industries. The color measurement instruments such as colorimeters and spectrophotometers, used for production quality control have many limitations. In many applications they cannot be used for a variety of reasons and have to be replaced with human operators. Machine vision has great potential for color measurement. The components for color machine vision systems, such as broadcast quality 3-CCD cameras, fast and inexpensive PCI frame grabbers, and sophisticated image processing software packages are available. However the machine vision industry has only started to approach the color domain. The few color machine vision systems on the market, produced by the largest machine vision manufacturers have very limited capabilities. A lack of understanding that a vision based color measurement system could fail if it ignores the basic principles of colorimetry is the main reason for the slow progress of color vision systems. the purpose of this paper is to clarify how color measurement principles have to be applied to vision systems and how the electro-optical design features of colorimeters have to be modified in order to implement them for vision systems. The subject of this presentation far exceeds the limitations of a journal paper so only the most important aspects will be discussed. An overview of the major areas of applications for colorimetric vision system will be discussed. Finally, the reasons why some customers are happy with their vision systems and some are not will be analyzed.

For the past three summers, students at the University of Michigan-Dearborn have been participating in the development and testing of various aspects of machine vision systems with support from the National Science Foundation under the Research Experiences for Undergraduates (REU) program. Much of the work has involved algorithm development since useful work can be performed with a fairly modest programming background. Benchmarking of various algorithms is a related activity that has seen much student participation. To a lesser extent, illumination and optics work has also been performed for the development of experimental setups and actual implementation of vision systems. Over the three-year duration of the program, a total of 34 students participated in these activities. While many of the participants were full-time students at the University of Michigan-Dearborn, others were from engineering colleges over a diverse geographical area. Summaries of a number of the projects is included here. It may be noted that the National Science Foundation has established the REU program to encourage more students to obtain advanced degrees in science and engineering and ultimately to pursue careers in research and development.