There are many challenges in inspecting inkjet nozzle plates. For example, orifice sizes have gotten smaller and smaller to support smaller drop sizes. Reflective plate materials create difficulties in capturing high-contrast, top-illuminated imagery. Time constraints in production environments require high-speed image capture, feature inspection and part disposition. Current systems use 2D CCD array cameras to step and repeat inspection along the length of each plate. Step and repeat methods have inherent limitations in inspection time, focus and illumination. In response to customer needs for a more robust methodology, a line-scan camera based inspection system has been developed. This paper will describe the line-scan based inspection system including considerations for image improvement through both hardware and software modifications. System capabilities and limitations will be discussed.

The authors present an application in quality control where a 100% inspection of the surface and shape of small stampings is done. The stampings have a size of approximately 1 inch in diameter and a complex disc-shaped geometry with free-form surfaces. Typical defects are various deformations like cutoffs, incomplete stampings, cracks and scratches, etc. on both sides of the parts. For the inspection of the two sides two independent sensor units with an extreme diffuse IR-LED flash illumination are used. The parts are turned back between the two inspection tasks by a specially developed turning arrangement. The image processing is based on an image subtraction approach and it is guided by an inspection plan to handle the part variants. The difference image between the actually controlled image and a reference image contains the potential defects. The reference image is computed from various master parts in a teaching process. With the help of a so called region image built from a CAD model of the stamping the defect pixels are sorted by the use of a clustering method according to their spatial appearance. The preprocessing and alignment of these images, the segmentation of the defects and the steps of the final decision are described.

Metallurgy industry which mainly transforms the steel or its derivative products into products with either better surface properties (thanks to the surface transformations...), or into different shape products (lamination...), involves some processing tools which can often generate flaws (cracks, grooves...) within the process. Prior to this study which concerns the control of laminated tubes, recipient for the uranium inside the nuclear reactor, the tubes were visually inspected after the lamination process. According to the quality estimation of the tube, subjectively done by an operator (between 1 and 4), the process was possibly stopped (grade 4). In order to obtain an objective control of the tube, a machine vision set-up was designed. The primary goal of this prototype is to provide a view of the whole surface of the tube to the operator.

This paper presents a novel machine vision system for quality control of wood planks. A general hardware configuration and suitable processing methods for detection and classification of biological and mechanical defects are proposed. High detection rates at required sped, low-cost and high degree of noise immunity are the main advantages of the proposed system.

In this paper, we describe the design and development of a dual-energy system used for x-ray screening of airport carry-on luggage. Dual-energy x-ray systems make it possible to measure the average atomic numbers of screened objects to enable their classification into three categories: inorganic, organic and mixed materials. Detection of organic materials, usually associated with dangerous compounds, mainly plastic explosives, is easier to achieve with dual- energy, as opposed to single-energy systems. The theory behind dual-energy systems is presented, followed by the design of a system based on a sandwich transmission detector arrangement with all its components, such as x-ray detectors, filter, operating tube, etc., and associated parameters are estimated according to simulation data. The process of generating the Z image, which includes the atomic number information from the two base images, is also described. After the prototype had been built, the unit was calibrated and images were taken with materials of known atomic number. Based on those measurements, the unit was tuned for optimal performance. Results comprise all decoding and compression tables for generating the images. Samples of Z and RO images taken from the unit are included and described in the report.

In this paper we present a technique which estimates the next angle dynamically, depending on the entropy of the silhouette actually acquired. The relation proposed guarantees a uniform object resolution on one side and a minimal number of acquisition steps on the other side. The method has been tested on synthetic and real data with reasonably good results. The paper concludes with a presentation of results and an outlook on future work.

This paper describes a method for automatic segmentation of facial features such as eyebrows, eyes, nose, mouth and ears in color images. This work is an initial step for wide range of applications based on feature-based approaches, such as face recognition, lip-reading, gender estimation, facial expression analysis, etc. Human face can be characterized by its skin color and nearly elliptical shape. For this purpose, face detection is performed using color and shape information. Uniform illumination is assumed. No restrictions on glasses, make-up, beard, etc. are imposed. Facial features are extracted using the vertically and horizontally oriented gradient projections. The gradient of a minimum with respect to its neighbor maxima gives the boundaries of a facial feature. Each facial feature has a different horizontal characteristic. These characteristics are derived by extensive experimentation with many face images. Using fuzzy set theory, the similarity between the candidate and the feature characteristic under consideration is calculated. Gradient-based method is accompanied by the anthropometrical information, for robustness. Ear detection is performed using contour-based shape descriptors. This method detects the facial features and circumscribes each facial feature with the smallest rectangle possible. AR database is used for testing. The developed method is also suitable for real-time systems.

In this paper we discuss the characteristics of pipeline features, e.g. welds, cracks and erosion objects, and present a method for the diagnostically lossless compression of Transverse Field Inspection data, a new type of pipeline inspection data in contrast to the traditional Magnetic Flux Leakage inspection data.

Quality control by artificial vision is getting more and more widespread within the industry. Indeed, in many cases, industrial applications require a control with high stability performance, satisfying high production rate. The purpose of this paper is to present a method to detect in real time defects located on the circumference of industrial parts that have a circular shape. Meanwhile, production steps can lead to an oval shape or to different sizes of parts. These two phenomena can lead to miss defects. Therefore a constant mask will not be able to detect these defects correctly. The control of the circularity of these parts can be achieved in two steps.

A True Color Tube Bore Inspection System (TCTBIS) has been developed to aid in the visual nondestructive examination of the inside of small diameter tubes. The instrument was developed to inspect for the presence of contaminants and discoloration inside the tube. The tubes, which have a 1.5 - 1.7 millimeter inside diameter, are integrally attached to pressure vessels that are filled to high pressure through the tubes. The latest version of the TCTBIS can operate in two modes. In the forward-looking mode a borescope is used to look down the length of the tube. In the side-looking mode, a tube containing a 45 degree(s) mirror is placed over the forward-looking borescope so that a direct view of the sidewall of the tube can be seen. The work reported here is a comparison of the relative performance of these two operating modes in terms of image quality and data acquisition speed. Each mode uses an entirely different method of image acquisition and unwrapped image reconstruction. These methods along with comparison results and suggestions for improvements will be discussed in detail.

We report on algorithmic aspects for the automated visual quality control for grading of brown eggs. Using RGB color images of four different views of every egg enabled to analyze the entire eggshell. The scene was illuminated using a set of white fluorescent tubes placed in a rectangular grid. After detection and approximation of the egg contour (ellipse fitted), the color was corrected to compensate for the elliptical shape of the eggs. A second order polynomial was fitted through points taken from subsequent horizontal lines inside the egg. Iteration was used to reject outliers (most likely points with visual defects). The shape- corrected intensity was calculated as the signed difference between polynomial and measured value, increased with the average egg intensity. Based on the corrected color, dirt regions like yolk, manure, blood, and red mite spots were segmented from the egg-background. Features based on color and shapes were calculated for every segmented region as the combined space and color moments of zeroth, first and second order. A classifier identified most of the defective eggs. Elimination of false rejects due to mirror reflection of the light tubes on some eggs (segmented because of the different color) is currently under investigation.

The appearance of skin colors in the images depends among other things, on the camera, the calibration of the camera, and the illumination under which the image was taken. In this study, we investigate how the skin colors appear in the chromaticity coordinates of different color spaces like HSV/HSL, normalized rgb, YES and TSL. For this purpose, we have taken images of faces under 16 different illumination/camera calibration conditions using simulated illuminants (Horizon, A, fluorescent TL84 and daylight) with different RGB cameras (1CCD web cameras and a 3CCD camera). In the making of this series of 16 images, first the selected camera was calibrated to one of the four light sources and an image was taken. After that the light source was changed to the other light sources and at each time the person was imaged. The process was repeated to the other two light sources. The same procedure was done for all four light sources and for each camera. The skin regions were extracted from these images and this skin data was then converted to different color spaces. We inspected how the chromaticities of different skin color groups in these color spaces overlap in images taken in all 16 different cases and only in those cases in which the selected camera was calibrated to the current illuminant. These investigations were also made between different cameras. In addition to this, we examined the overlapping of all skin chromaticities from the different skin color groups between cameras.

Intramuscular fat content in meat influences some important meat quality characteristics. The aim of the present study was to develop and apply image processing techniques to quantify intramuscular fat content in beefs together with the visual appearance of fat in meat (marbling). Color images of M. longissimus dorsi meat samples with a variability of intramuscular fat content and marbling were captured. Image analysis software was specially developed for the interpretation of these images. In particular, a segmentation algorithm (i.e. classification of different substances: fat, muscle and connective tissue) was optimized in order to obtain a proper classification and perform subsequent analysis. Segmentation of muscle from fat was achieved based on their characteristics in the 3D color space, and on the intrinsic fuzzy nature of these structures. The method is fully automatic and it combines a fuzzy clustering algorithm, the Fuzzy c-Means Algorithm, with a Genetic Algorithm. The percentages of various colors (i.e. substances) within the sample are then determined; the number, size distribution, and spatial distributions of the extracted fat flecks are measured. Measurements are correlated with chemical and sensory properties. Results so far show that advanced image analysis is useful for quantify the visual appearance of meat.

This paper presents a model-based algorithm for localization and measurement of SMC objects in miniature scale. The algorithm comprises a coarse-to-fine search strategy that fully utilizes all available information from object region and contour. In the coarse search stage, the moment transformations combined with an iterative segmentation scheme enables object localization performed in large search space and produces rough estimates of the pose parameters. Then, the canny edge detection and interpolation process is applied to obtain accurate object boundary as best as possible that allows an iterative optimization procedure to refine the initial estimates. The essential idea underlying our approach is by modeling the object with a polygonal shape and assumes that the input object is located on simple image background, which is suitable for a pick-and-place system used by semiconductor assembly. Finally, the result shows a compromise between accuracy and image resolution.

A novel technique that can measure the angle of a vertical microstructure with an accuracy of 0.05 degree(s) has been developed. The microstructure has a movable micromirror that is the main component of Micro-Optical Cross Connect (MOXC) that switches and couples optical signals in the optical communication networks. Since the performance of MOXC strongly depends on the angle between the vertical micromirror and the substrate, it is very important to measure the angle with high resolution better than 0.05 degree(s). In order to get the side view of the microstructure, a microscope imaging system that employs a long working distance objective lens and CCD camera has been built.

We are developing a sensor camera system for automatically tracking and determining the positions of subjects moving in three-dimensions. The system is intended to operate even within areas as large as soccer fields. The system measures the 3D coordinates of the object while driving the pan and tilt movements of camera heads, and the degree of zoom of the lenses. Its principal feature is that it automatically zooms in as the object moves farther away and out as the object moves closer. This maintains the area of the object as a fixed position of the image. This feature makes stable detection by the image processing possible. We are planning to use the system to detect the position of a soccer ball during a soccer game. In this paper, we describe the configuration of the developing automatic tracking sensor camera system. We then give an analysis of the movements of the ball within images of games, the results of experiments on method of image processing used to detect the ball, and the results of other experiments to verify the accuracy of an experimental system. These results show that the system is sufficiently accurate in terms of obtaining positions in three-dimensions.

Image alignment is the most crucial problem in industrial visual inspection. Traditional intensity-matching based methods, including the normalized correlation method, are not robust against non-uniform illumination variations. In this paper, we present a generalized intensity-based matching approach to accomplish accurate and robust image alignment under high-level noises and large non-uniform illumination variations. This generalization is an extension of our previous FLASH (Fast Localization with Advanced Search Hierarchy) algorithm for image alignment.

The ultimate goal of vide compression is to maximize picture quality while minimizing bandwidth requirement. In most video storage and communication applications, the constraints are often expressed as limitation in delivery bandwidth or storage capacity. The objective of video encoder development is, therefore, to minimize the amount of distortions introduced by video compression and transmission. As an enabling technology, measurement of compression distortions and quality impact to end-users is crucial to video encoder optimization. While the field is developing quickly, there have been two different paradigms for video quality measurement that are being studied, picture quality models that use reference pictures, and the models that do not. An issue common to picture quality measurement in both paradigms is to obtain accurate measurement of picture distortions. In this paper, we review the requirements of these two measurement paradigms and propose two image analysis methods that address some specific issues of picture distortion measurement. First we describe a fast video alignment approach necessary for picture distortion measurement models that require references. In the rest of the paper, we propose a blur estimation scheme to measure blurring degradation introduced by video compression and imaging systems. We will then review its reference-free distortion measurement performance using data from two experiments.

This paper presents the use of simple PC based tactile sensing system for object recognition. Software has been developed to acquire, process and analyze the tactile images. The recognition is based on the use of invariant moments and a back propagation neural network. The software also provides information regarding contact area, height, position and orientation of the object. The paper deals with details of the software developed. Various experiments carried out to test the performance of the system have also been described along with the results obtained.

Object identification by matching is a central problem in computer vision. A major issue that any object matching method must address is the ability to correctly match an object to its model when only a partial view of the object is visible due to occlusion or shadows (or any other reason). In this paper we introduce surface boundary signatures as an extension to our surface signature formulation. Boundary signatures are surface feature vectors that reflect the probability of occurrence of a surface boundary feature.

When the analog-to-digital mapping to be learned by any pattern recognition scheme satisfies a certain PLI condition, a one-layered, hard-limited perceptron (OHP) is enough to be used for recognizing any unlearned patterns with high robustness. Generally, the PLI condition is satisfied for most practical pattern recognition applications. When this condition is satisfied, then an automatic feature extraction scheme can be derived from an N-dimension geometry point of view. This automatic scheme will automatically extract the most distinguished parts of the pattern vectors used in the training. It selects the feature vectors (sub-vectors of the pattern vectors) automatically according to the descending order of the volumes of the parallelepiped spanned by these sub-vectors. Theoretical derivation revealing the physical nature of this process and its effect in optimizing the robustness of this novel pattern recognition system will be reported in detail.

Specular surfaces inspection is a problem met frequently within the automatic control of metallic products. A tried technique to reveal the aspect defects is the imaging of the reflection of a structured lighting as Dark Field Illumination through the surface. In order to inspect the whole surface, an element of the lighting structure has to scan every part of the surface. In the case of important surface curvature gradients, entire scanning is not ensured if the object is moving in front of the static lighting. To overcome this limitation, an inverse process is proposed: the lighting structure is dynamic while the object is static. The scanning of the surface by the various lighting configurations enables the aspect control. A modeling of the surface to be inspected and defects to be detected is made. Inverse ray tracing is used to analyze the reflection of the lighting through the surface. This modeling enables to ensure the revealing of defects by respecting the chosen criteria of detection. A relation binding the physical size of defects and its size on the image is established. A metrological approach of the problem is then performed.

This work presents an automated shape inspection system for 2D objects with variable luminance. The system installed in the steel industry, captures linear images of plates at high temperature (700 - 1200 degree(s)C) while they are moving on a roll path. The main objective of the system is to capture the shape of the head and tail of the plates. These shapes are used to optimize the rolling parameters of the plate mill in order to minimize waste. The radiation generated by the plates in the visible and infrared zones of the spectrum (largely dependent on their temperature) is directly captured by the linear camera of the system with no additional artificial illumination. While most of the research work has been focused on obtaining the optimal illumination for the objects inspected, this work deals with the particular case of objects which irradiate their own light. This system automatically adapts itself to acquire images of plates with different levels of luminance using a mechanisms that calculates the proper exposure time to acquire each image. The mechanism integrates two basic actions: a feedforward control and an adaptive feedback control loop.

In this paper we propose a system for capturing and processing car identification labels. The system captures the required images for the registration process using a CCD camera with an appropriate mount. Images are analyzed and stored by the system, generating a report, where all the captured visual information is included. The information is further stored in a way, that allows the retrieval of each report. Results and discussion about the system are presented.

An image entails a huge amount of data and information. For this reason, image synthesis and analysis by computer systems requires a high processing time. This represents a handicap in systems where real time processing or an immediate interpretation is demanded as in visual inspection industrial applications. Present work, introduces a computer architecture for the construction of a compact real-time system for high speed visual inspection. The vision system is essentially a Cellular Neural Network Computer (CNN-C) basically composed of a Cellular Neural Network Universal Machine (CNN-UM), an analog memory, an imager and a control unit with mixed-signal properties. This prototype has some limitations, but represents the first approximation of a new kind of systems for visual inspection. The CNN-C prototype will be tested in visual inspection of paper, metal and polymer surfaces. Besides the CNN-C can be used in many other image processing tasks, such as coding, singularity detection or multiresolution representation.

This paper presents a modified Hough transform (HT) by formulating the conjugate pair for line detection. By considering conjugate pair of the HT, a fast computing algorithm can be derived. The concept of conjugation is applied to the Radon transform, a generalized HT, and the Gabor transforms. Formulation of the conjugate pairs of 3D Hough transform is also presented.

We propose a new approach to building appearance models of 3D objects which is based on Canonical Correlation Analysis (CCA). In appearance based modeling, instead of building an explicit object model (e.g., 3D geometrical object model), a low dimensional object representation is obtained from a set of images. In standard appearance models typically Principal Component Analysis (PCA) is used for feature extraction. In our experiments we compare the performance of standard appearance models based on PCA and models based on CCA for 3D pose estimation. Results indicate that, while getting by with a smaller number linear features, CCA-based models perform consistently better.

Analog CNN array computer arises as an alternative to traditional digital processors in many industrial inspection like visual quality control of metal laminants, capable of make in a single chip Tera equivalent operations per second. A 4096 analog CNN processor array is able to perform complex space-time image analysis, being much faster than a camera- computer system in continuous inspection applications. Both chips have been implemented in CMOS technology and they are managed by a 32-bit high-performance low-cost micro- controller that closes the pan, tilt, lighting, focus and zoom loops required in the implementation of the active vision strategies. Several convolution masks for the Cellular Processors has been selected to detect particular changes in the texture, size, direction or orientation of the image entities, reprogramming `on the fly' the pixel resolution of shape when necessary. Laboratory results present these Cellular Processors and multiple resolution imager circuits as a promising architecture for visual inspection of industrial processes in real time.