We propose a content-based summary generation method using MPEG-7 metadata. In this paper, the important events of video are defined and subsequently shot boundary detection is carried out. Then, we analyze the video contents in the shot with multiple content features using multiple MPEG-7 descriptors. In experiments with a golf-video, we combined motion activity, edge histogram and homogeneous texture for the detection of event. Further, the extracted segments and key-frames of each event are described by XML document. Experimental result shows that the proposed method gives reliable summary generation with robust event detection.
A two-class classification method of image patterns using principal component analysis (PCA) is proposed, in which classification is performed in the two-dimensional (2-D) space constructed by the reconstruction errors. The reconstruction error is computed using PCA for each assumed class. Training data sets are used to compute eigenvectors with which PCA reduces the dimensionality of the input vector space and reconstructs an input vector in the reduced space. The line equation with two parameters is defined as a linear decision boundary and these parameters are estimated by probabilistic approach. Also its application to face detection is experimented.
This paper presents a new method of fitting geometric objects to measurement data. The coefficients of the implicit equations describing the object to be fitted are used as the Grassmann coordinates. The design matrix is then formed as the dual space to the Grassmann coordinates. The analysis of the null space of the design matrix yields the solution being sought. Degenerate data or models can be identified by analyzing the Grassmann manifolds in the null space. The method can also be applied to fitting coupled geometric objects, such as concentric circles. A solution has been found which enables the linear fitting of rational polynomials a problem which was considered to be non-linear. The method is demonstrated for conic sections, quartics, rational Bezier curves and for some examples in image processing.
The human visual system seems very powerful when it is a question to identifying an object or a portion of an object in movement, such as a textured surface moving in a 3D textured environment. In such a situation, the visual impression of an observer depends on many factors, including the nature of relative movement between the scene and the observer, the kind of lighting and the surface aspect of the plane being studied. In this paper, we propose a method of characterization of textured surfaces moving in a 3D scene. The various analyzed images are strongly textured, and do not necessarily include periodic elementary patterns. The autocorrelation function associated with an optical model with the system scene-camera under the hypothesis of a weak perspective projection is used. We use the fact that the autocorrelation function of an image and of its affine transformed version are related by this transformation. 3D-rotations of the textured plane are studied by means of Euler's angles. On a set of 1008 synthetic images, the accuracy obtained for the angles of rotation is characterized by a standard deviation of about 9 degrees. It attains 4.1 degrees on a small database of real images.
Camera ego-motion is computed utilizing optical flows which are obtained from sparsely located feature points, where the corresponding points are determined through the information fusion of the correlation and the system model-based prediction. Since the ego-motion of the camera is obtained utilizing optical flows, the accuracy of recognized motion depends highly on the correctness of optical flows. Therefore, the ego-motion utilizing feature-based optical flows are more reliable than those of other points due to the distinctive characteristic of feature points. The technical bottleneck of this category of solution is the matching of corresponding points. In this paper, the correlation and prediction is fused to determine trustable matching pairs. For better prediction, system dynamic model is employed. The effect of the proposed algorithm has been shown through the motion estimation of the camera installed on a dynamic system.
Non-intrusive inspection and non-destructive testing of manufactured objects with complex internal structures typically requires the enhancement, analysis and visualization of high-resolution volumetric data. Given the increasing availability of fast 3D scanning technology (e.g. cone-beam CT), enabling on-line detection and accurate discrimination of components or sub-structures, the inherent complexity of classification algorithms inevitably leads to throughput bottlenecks. Indeed, whereas typical inspection throughput requirements range from 1 to 1000 volumes per hour, depending on density and resolution, current computational capability is one to two orders-of-magnitude less. Accordingly, speeding up classification algorithms requires both reduction of algorithm complexity and acceleration of computer performance. A shape-based classification algorithm, offering algorithm complexity reduction, by using ellipses as generic descriptors of solids-of-revolution, and supporting performance-scalability, by exploiting the inherent parallelism of volumetric data, is presented. A two-stage variant of the classical Hough transform is used for ellipse detection and correlation of the detected ellipses facilitates position-, scale- and orientation-invariant component classification. Performance-scalability is achieved cost-effectively by accelerating a PC host with one or more COTS (Commercial-Off-The-Shelf) PCI multiprocessor cards. Experimental results are reported to demonstrate the feasibility and cost-effectiveness of the data-parallel classification algorithm for on-line industrial inspection applications.
The optical pickup of a magneto-optical drive is constructed of millimeter-size optical components, including a laser diode (LD), a collimating lens (CL), a polarizing beam splitter, some kinds of prisms, and photo detectors (PD). Each component must be assembled with micrometer-order accuracy. In particular, the astigmatism, which is adjusted by changing the position of the CL, must be within 0.65 micrometers in the optical axis direction. To enable the CL position to be adjusted quantitatively, we developed an adjustment method that uses passive alignment. We estimated the astigmatism by analyzing an image of the LD, which is acquired through the CL by illuminating the LD with the polarized light. We developed an optical-pickup adjustment system using the proposed method and tested its effectiveness experimentally. The results showed that this system can be used to adjust the accuracy to within 0.65 micrometers. Because the image of the LD and the PD are acquired clearly by this system, it should be useful for not only adjusting the optical pickup but also for visually inspecting the LD and PD.
Removing speckle noise of electronic speckle pattern interferometry (ESPI) from a single ESPI fringe pattern while keeping the fringe feature is a difficult problem and remains unsolved. In this paper, the spin filtering proposed by the authors is developed further with curved surface windows to filter off speckle noise from a single speckle fringe pattern. With the new filtering, the speckle noise can be removed nearly completely from a single speckle fringe pattern. The most important advantage of the method is that no blurring effects occur for speckle fringe patterns and the smooth fringe pattern of a phase field is retrieved and derived.
In this paper, a new target extraction algorithm is proposed, in which the coordinates of target are obtained adaptively by using the difference image information and the optical BPEJTC with which the target object can be segmented from the input image and background noises are removed in the stereo vision system. The proposed algorithm, firstly, extracted the target object removing the background noises through the difference image information of the sequential left images and then, control the pan/tilt and convergence angle of the stereo camera by using the coordinates of the target position obtained from the optical BPEJTC between the extracted target image and the input image. From some experimental results, it is found that the proposed algorithm can extract the target object from the input image having the background noises and then, effectively track the target object in real-time. Finally, a possibility of implementation of the adaptive stereo object tracking system by using the proposed algorithm is also suggested.
A vision system for the automatic quantification of fabric geometric distortion has been implemented and tested. The intended utility of this system is to replace the manual measurement of fabric shrinkage or growth as governed by the AATCC (American Association of Textile Chemists and Colorists) Test Method 135. In the near future, other capabilities, such as automatic quantification of fabric smoothness, will also be incorporated. The system uses commercial, off-the-shelf hardware components, together with a customized image processing algorithm to capture digital images of pre-marked fabric swatches and to accurately measure the distance between the benchmarks before and after laundering. The primary focus of this paper is a description of the algorithm that detects these benchmarks. This robust algorithm detects the marks without regard to: (1) changes in the texture or the color of the swatches, (2) inter-fabric changes in the benchmark colors, (3) changes in the fabric contrast due to scanning or laundering, (4) presence of noise, or (5) slight rotations of the swatches during scanning. The presented system has been under routine testing at the International Textile Center of Texas Tech University, as well as the laboratories of Cotton Inc., with the computed dimensional changes and the manual measurements possessing a nearly perfect linear correlation.
The aim of this research was to study a system of acquisition and processing of images capable of confronting colored wool with a reference specimen, in order to define the conformity using objective parameters. The first step of the research was to comprise and to analyze in depth the problem: there has been numerous implications of technical, physical, cultural, biological and also psychological character, that come down from the attempt of giving a quantitative appraisal to the color. In the scene of the national and international scientific and technological research, little has been made as regards measurement of color through digital processing of the images through linear CCD. The reason is fundamentally of technological nature: only during the last years we found the presence on the market of low cost equipment capable of acquiring and processing images with adequate performances and qualities. The job described has permitted to create a first prototype of system for the color measuring with use of CCD linear devices. -Hardware identification to carry out a series of tests and experiments in laboratory. -Verification of such device in a textile facility. -Statistics analysis of the collected data and of the employed models.
We propose and experimentally demonstrate a fluorescent imaging technique for surface quality control of wet-cleaned silicon wafers. This simple technique allows macro- and microscopic imaging. Submicron resolution and fast scanning are successfully demonstrated. Distribution of water stains is measured using this novel technique and correlated to the surface structure.
We have developed a self-organizing map (SOM) -based approach for training and classification in visual surface inspection applications. The approach combines the advantages of non-supervised and supervised training and offers an intuitive visual user interface. The training is less sensitive to human errors, since labeling of large amounts of individual training samples is not necessary. In the classification, the user interface allows on-line control of class boundaries. Earlier experiments show that our approach gives good results in wood inspection. In this paper, we evaluate its real time capability. When quite simple features are used, the bottleneck in real time inspection is the nearest SOM code vector search during the classification phase. In experiments, we compare acceleration techniques that are suitable for high dimensional nearest neighbor search typical for the method. We show that even simple acceleration techniques can improve the speed considerably, and the SOM approach can be used in real time with a standard PC.
A valuable visual indicator to grade the stiffness and strength of planks can be obtained by analyzing the structure of the grain on it. To integrate an analyzing image vision module in an industrial selection process a real-time system is needed to build. Two main objectives must be reached: First a stable edge detector should extract the grain edges. Second these grain edges have to be tracked to achieve a complete grain representation. This representation can be used to analyze the regularity of the grain. Since the visual nature of grain varies a lot even on a single plank we present an edge detector which is adaptive and a grain tracking algorithm capable of closing gaps between pixels. Both steps work in real-time (i.e. 5 frames per second resulting in 1 meter per second).
In this paper, an efficient tunnel crack detection and recognition method is proposed. It combines the analysis of crack intensity feature and the application of Support Vector Machine algorithm. At first, the original image is transformed into a binary image. Based on two thresholds technique, the object edge image can be obtained. Then assuming the image can be separated to some local images, we catagorize the local image into three types of pattern. They are the crack, non-crack and intermediate type, which have both of the two properties. A trainable classifier is built to classify these patterns. During this process, Balanced sub-images that satisfy for the two centers of geometric and gravity, are used as a trainable sample for the classifier. This leads to an effective classification system.
Car road holding is linked to many factors including adherence, which is strongly related to the roughness of the road coating and of its evolution under the effect of traffic. Traffic induces a progressive wear of the road surface which results in a modification of its local relief and of its roughness. We study road coverings in order to know the variations of their wear levels over time by analyzing the micro texture of these road images. The transcription of roughness criteria in image analysis requires, on one hand, the development of a photometric model for the coating surface, and, on the other hand, a modeling of the profile of the road coating. The method suggested in this article is based on a photometric model of the surface of the coating from which we study the statistical properties of : the distribution of the gray levels in the image, the distribution of the absolute value of its gradient, and the form of its autocorrelation function. Experiments have been done with images of road coverings at different wear levels. The obtained results are similar to those obtained by a direct method of contact measuring.
The assembled PCB is made to insert the electric components manually or automatically. After inserting some electric elements on a PCB, the PCB is processed in the soldering process. However, there are some insertion or mounting errors in the PCB after the inserting or soldering process. The machine vision has been used to detect these errors. When a CCD camera and an X-Y positioning table in a machine vision use, it takes a high cost and long inspection time because it should move on some inspecting spots. In case workers inspect a PCB with eyes to find the errors, a result is different depending on worker's physical and mental condition. To solve these problems, we used a PC, universal serial bus (USB) hub to use several USB devices and several USB cameras that were located over the PCB and inspected whether the errors were found or not. In this study we found the fact that our system's error was lower than worker's and it took less cost than other machine vision system using the CCD camera and X-Y table.
This paper shows an efficient and reliable method for the detection of surface defects with a three dimensional characteristic whereby the surface reflection properties are altering strongly. Due to this fact traditional intensity imaging techniques yield inferior performance. Therefore, light sectioning in conjunction with fast imaging sensors is applied to gather the range image of the steel block. Two different approaches for defect detection are treated, whereby the first algorithm is based on a line-wise examination of the acquired profiles by unwrapping the surface using spline interpolation. The noise in the unwrapped orthogonal distance may be reduced by applying statistical measures. The second method refers to surface segments and is based on the mean square error between the segment and its approximation gained from singular value decomposition. Due to vibrations the acquired profiles are arbitrary located within a range of a few millimeters which requires a geometric transformation to reconstruct the three dimensional surface of the steel block.
The topic of this research is to the study the feasibility of a machine vision prototype for the control of metallic tubes (used in water pomp). Nine different kinds of defects located everywhere on the tube have to be detected: The defects are: on the top: little hollows, bumps, and excesses of material on the body: horizontal bumps, vertical bumps, vertical scratches and finally on the bottom: vertical ridges, holes, and bumps. As the defects on the top of the tube are very small, a grazing angle is used to light the tube. The camera is set on the opposite side of the tube with the same angle. Hollows and bumps are both detected by a vertical Sobel gradient. For the third defects, the excess of material projects its shadow on the top of the tube, and defects are detected by looking for a dark region instead of a lighted one. To inspect the rest of the tubes, a neon tube with a diffuser is employed to homogeneously light the body and the bottom of the tubes. Association of gradient operators, threshold procedures enables to find all the defects.
Metallurgy Industry which mainly changes 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 generate flaws (cracks, grooves...) within the process. Prior to this study the laminated tubes, recipient for the uranium inside the nuclear reactor, where visually inspected after the lamination process. According to the quality estimation of the tube, subjectively done by the operator (between 1 and 4), the process was possibly stopped (grade 4). In order to have a more objective control of the tube, a machine vision set-up was developed. The primary goal of this prototype is to provide a view of the whole surface of the tube to the operator. Various lighting systems where tested so as to reveal the maximum number of defect (tool marks, scratches, ..). The tube surface being not even (rough), we found that an homogenous lighting of the scene enabled a clear inspection of the tube surface and reveal most of the defects. Unfortunately a structured lighting system was also required for the tool-marks to be visible. Then, different image processing tools have been applied to the images. At this point, all the defects are detected. Further experiments are currently being done to classify the defects.
In this paper is presented a application of image processing algorithms to fabrics inspection. An experimental apparatus was developed in order to obtain good color images to be processed. After converting this images to gray levels, it is possible to obtain a binary image containing the defect, after applying threshold segmentation, statistical analysis and mathematical morphology. The automatic system needs to be trained in order to characterize the fabric using the mean and standard deviation of it's gray levels. Analysis were made with four types of defects: gap; broken thread; bulky thread; stain. To detect color variations in the fabrics, a novel approach for color segmentation is presented. Finally an application was developed using MicrosoftTM Visual C++, based on object oriented programming, to achieve automatic fabric inspection by machine-vision.
In the textile industry, the quality of the finished fabric is subjectively determined by human inspectors resulting in inconsistent quality control. This paper presents a system and a methodology that detects the defects on textile fabric automatically and classifies them into one of the pre-defined classes with repeatable accuracy. The overall inspection process is based on color image processing. Emphasis is given on a problem of feature extraction using different color spaces. For these purposes the first order and the second order statistics are implemented. Since both the calculation and comparison of the features consume time, the template matching and k-NN classifier are used for classification.