INTELLIGENT AND OPTIMAL NORMALIZED CORRELATION FOR HIGH- SPEED PATTERN MATCHING. Abstract
|
|
- Stuart Stevenson
- 5 years ago
- Views:
Transcription
1 INTELLIGENT AND OPTIMAL NORMALIZED CORRELATION FOR HIGH- SPEED PATTERN MATCHING Swami Manickam, Scott D. Roth, Thomas Bushman Datacube Inc., 300, Rosewood Drive, Danvers, MA 01923, U.S.A. Abstract The vision industries have used normalized correlation to reliably locate patterns with high accuracy. However, normalized correlation is computationally very expensive. Researchers were able to reduce the computational complexity of normalized correlation by using data pyramids and thus making it find patterns in real-time, but only in spatially translated images. Any attempts to extend normalized correlation to search for a rotated and scaled pattern proved to be too slow for most vision applications. Other attempts by vision industry researchers to speed up normalized correlation, such as skipping pixels, deteriorated the performance. In addition, today s vision industries need to handle nonlinear changes in brightness, process variations such as multilayer buildup in wafer production, blurring, and perspective distortion. In this paper, we discuss an algorithm that has been developed to resolve all these issues. The developed algorithm extracts only the essential information from patterns through a training process built upon normalized correlation. Using an intelligent- search technique and normalized correlation as the matching criteria, it locates the positions, orientations, and scales (both horizontal and vertical) of one or more trained patterns in real time, while maintaining high accuracy. Overview Registration of an image with respect to a reference pattern has a wide range of applications such as: Wafer alignment, using arbitrary artwork as the pattern to recognize. Fiducial recognition for PCB assembly by a pick-and-place robot. Registration of alignment marks on printed material to be inspected (e.g., wallpaper rolls, currency sheets). Robot vision guidance, locating objects on conveyor belts, pallets, and trays. Vision industries and researchers have employed a mathematical procedure known as normalized gray-scale correlation (NGC) to locate the reference pattern within an image under consideration. In this method, the reference pattern is shifted to every location in the image, the values are multiplied by the pixels that are overlaid, and the total is stored at the position to form an image showing where regions identical or similar to the reference pattern are located. In order to normalize the result of this pattern matching or correlation without the absolute brightness value of the region biasing the results, the operation is usually calculated as the sum of products of the pixel brightness divided by their geometric mean (Russ, 1992).
2 Over the years, NGC has been proven to be a robust and reliable method (Rosenfeld et al., 1982). If sub-pixel accuracy is required, the correlation surface may be interpolated and accuracy of better than 1/16 th of a pixel can be achieved (Gleason et al., 1990). This conventional NGC is invariant to linear changes in brightness, but that is where its invariance ends. In other words, conventional NGC has very little tolerance to any changes in rotation, scale, perspective distortion, non-linear changes in brightness, and multilayer buildup in wafer manufacturing. In addition, NGC is computationally very expensive (O(n 4 ) to find translated patterns). The formula for calculating the NGC score NC is: NC = ( I I )( T T) x, y ( I I ) ( T T) x, y x, y 2 2, where I = I N and T = T N T and I are template and image data, respectively, and N is the number of template pixels. To speed up NGC, Datacube and others have used pipeline image processing. While this allows more pixels to be processed per second, the computational cost of NGC remains the same (O(n 4 ) to find translated patterns). Several strategies such as skipping every other pixel and using random pixels have been used to reduce the number of pixels processed. But these strategies significantly deteriorate the reliability of correlation. Another frequently used strategy is use of image pyramids. An image pyramid is a succession of images where each image is down-sampled from its predecessor. In this approach, a coarse match is found at the top of the pyramid and some sort of hillclimbing strategy is utilized to traverse through the successive pyramid images. This significantly reduces the number of pixels used in correlation. With pyramiding, NGC is able to find patterns in real time, but with variance only in translation. However, in many applications, the given template and the observed image are not only spatially translated but are also relatively rotated and may be even scaled. In such cases, the NGC maximum of a 2-D pattern has to be searched in a 4-D parameter space: translation X, translation Y, rotation, and scale. This can become quite impractical (O(n 6 )) unless reasonable estimates of the scale and rotation are given (Jain, 1989). Today s vision industries require a pattern matching algorithm that can quickly locate a pattern that may have been rotated, scaled, blurred, occluded, distorted due to perspective, and illuminated differently (exhibiting non-linear change in brightness). Instead of attempting to adapt the reliable NGC to these new requirements by intelligent optimization, researchers have moved onto other techniques such as geometry-based, pattern-recognition algorithms. In this paper, we discuss an intelligent optimization of NGC to address these new requirements. The resulting algorithm finds the patterns faster and more accurately than the new geometry-based, pattern-recognition algorithms. The developed procedure, known as vsfind in its implemented form, consists of two algorithms:
3 1. Automatic training, during which the template is distinguished from other objects and the background within the search space, and 2. Run-time pattern recognition, during which the trained template is located in a new image. Development of an Intelligent and Optimal NGC Adaptation The major weakness of NGC is its computational complexity. Since NGC computes dot products, its computational complexity is directly proportional to the template size and image search area. Speeding up of the NGC can be achieved only by reducing the number of data points involved in the correlation. However, care must be taken not to degrade the accuracy and reliability of NGC. Do all pixels contribute equally or do some pixels contribute more than the others to the correlation score? If the latter is true, then how do we identify the influential pixels? In 1994, Krattenthaler et al. analyzed correlation-based template matching and reported that not all pixels contribute equally in correlation. They proposed Point Correlation, where matching is performed with a pre-computed set of points from the reference pattern or the template. In their method, a set of correlation-sensitive points (points of the template with the greater influence on template matching) is selected in a training session. Intuitively, the selected set of points corresponds to the most important features of the template, knowing the set of possible template transformations (translation, rotation, scaling, etc.). To determine this crucial set of points, they proposed an iterative algorithm as follows: 1. Compute a point set P M with M initial points (randomly select three points, preferably on edges). 2. Iteration step Assume a point set P L consisting of L points, where L M has already been computed. Then find the new set P L+1 in the following manner: a. For each point X j = (x i,y i ) in the template with X j P L Compute the correlation result R j (i) for all transformations i, 1 i N, using point correlation with the set of points P L X j. Compute a correlation measure CM j of the correlation result R j (i) that determines the quality of the point X j. b. Choose the point X j to be the new element of P L+1, whose correlation score CM j is a maximum. This technique clearly maintains the NGC s strength of accuracy and reliability by selecting the correlation-sensitive points. By reducing the number of points used in the correlation (from O(n 2 ) to O(n)), the run-time computations are reduced from O(n 6 ) to O(n 5 ). However, the training procedure for selecting such a set of points is computationally very expensive. For example,
4 Let the template be of size n x n pixels. Let the number of translations possible be of the order of n x n. Let the possible combinations of rotations and scale be O(n 2 ). Then N in step 2.a would be of O(n 4 ). Computing the correlation score for a single transform involves O(n) multiplies and additions. To select the most influential template pixel, all of the not-yet-selected template pixels (O(n 2 )) are considered in each iteration. Therefore, the number of computations required to select one pixel would be O(n 4 ) * O(n) * O(n 2 ). To select n pixels, the required number of computations would be: ( n) * O(n 7 ), which is O(n 9 ). Even with modern computers, training time on the order of O(n 9 ) is impractical for the vision industries, where both the run-time and training speed are important. Krattenthaler et al. did not propose any criteria to terminate the training except to set the number of points to some fixed value. The number of correlation-sensitive points of a template depends on the complexity of the template and the range of transformations (rotation, scaling, etc.). Hence, a criterion to determine the size of the set of points is also important in making the training practical. This point-correlation technique has the potential to meet some of the requirements (rotational and scale invariance, accuracy and reliability) of the vision industries. However, the computational complexity of the training makes the application of this technique impractical. The following section describes a procedure that optimizes the training through additional data-reduction techniques and patent-pending algorithms. Optimization of point-correlation training To reduce the training computational complexity, the data-reduction strategies are: 1. Data pyramiding is an independent data-reduction approach and can be combined with the point correlation. Pyramiding reduces both the template and image data approximately from O(n 2 ) to O(n). 2. The pyramiding concept can be extended to the transformations also. Use coarser rotational and scale transformations for the reduced data (top of the pyramid) with gradually finer transformations at the successive pyramid depths. This reduces the number of transformations approximately from O(n 2 ) to O(n). 3. Data pyramiding reduces the data with loss of details. Hence, the points selected at the reduced resolution must be tolerant to the set of coarser transformations. Patent-pending rules can be applied to select transformationtolerant pixels. 4. A heuristic (patent-pending) can be applied to the procedure of selecting the most influential pixel without compromising the optimality of the pointcorrelation template matching (Krattenthaler et al.). In addition, other calculations determine an optimal number of points for a given template and the transformations set.
5 5. Points selected at the spatially reduced resolution are extrapolated for the successive, higher resolutions. At the full resolution, additional points (a prespecified number based on the desired accuracy) are selected using the same strategy (#4, above) to achieve the full accuracy of the patternmatching algorithm. With these optimizations, the training computational complexity has been reduced from O(n 9 ) to O(n 3.5 ), which is practical. Fig. 1. below shows the points selected with the training algorithm for concentric squares with and without rotational transformations. The points selected at the spatially reduced resolution (and extrapolated for illustration) are shown as black x. The additional points selected at the full resolution to improve accuracy are shown as white +. With rotational transformations, the points selected by the training algorithm cluster around the corners, since corners are good indicators of the orientation of a square-shaped template and they are the furthest from the centroid of the template, thereby providing higher angular accuracy. Without rotational transformations, the selected points are spread out along all the edges. With rotation Without rotation Fig. 1. Points selected by vsfind s training engine to recognize a template of concentric squares with and without rotational transformations. Fig. 2. below shows the points selected to distinguish a ring from a disk with the same diameters. Since only the interior pixels corresponding to the hole are necessary to distinguish the two circular objects, the training algorithm selected only those points at the top of the pyramid. Points along the edges are selected at the full resolution to improve accuracy.
6 Fig. 2. Points selected by vsfind s training engine to distinguish a ring from a disk. Run-time pattern matching algorithm During run time, an exhaustive search is performed with the selected set of points representing the template and a coarse set of rotation and scaling transformations on the spatially reduced image data. Computed correlation scores are sorted, and the location, orientation, and scale that resulted in the best score are refined further for a possible match at the successive data pyramids. A hill-climbing approach is employed at each level of the data pyramid to converge toward the best score. At the full spatial resolution, another hill-climbing approach is utilized to achieve sub-pixel accuracy in X, Y, orientation, and scale. Point correlation combined with data and transformations pyramiding is capable of finding a trained template from an image with O(n 2.5 ) computations (translations: O(n), rotation and scale transformations: O(n), and the number of template points used in correlation: O(n 0.5 )). In addition, the multimedia extensions (e.g., MMX on Pentiums) of today s computers help to speed up repetitive multiplies. As a result, almost real-time performance is achieved. Achieving Other Requirements The vision industries need a pattern-matching algorithm that can handle nonsquare pixel data, perspective distortion, blurring, and nonlinear change in brightness. The following sections describe vsfind s enhancements to handle these. Perspective distortion and non-square pixel data Calibration is recommended before training and run time. The benefits of calibration are: 1. Pixels do not have to be square. 2. A template can be trained on a full, interlaced image and then later recognized in an image field. 3. Perspective distortion and skew are corrected, increasing accuracy. Fig. 3. illustrates the effect of perspective distortion on shape. A transformation that includes rotation, scaling, and correction for perspective distortion combined with
7 bilinear interpolation is applied to the sparse template during the search and final refinement for high accuracy. Object With Perspective Fig. 3. Effect of perspective distortion on shape. Recognizing blurred objects Blurring occurs when the object scene is out of focus or the depth of field is narrow and objects vary in height. The typical effect of blurring is the loss of details, like the output of a low-pass filter. Since the image pyramids are built via averaging, the points selected at the spatially reduced resolution are inherently tolerant to blurring. In addition, NGC s use of gray-scale data tolerates gradual variations such as smoothing due to blurring. vsfind versus geometry-based finders Machine vision is a crucial element of wafer manufacturing, for alignment and inspection. Contrast can change nonlinearly and unpredictably during the manufacturing process of a wafer. By operating on the edges of the image instead of the raw grayscale data, NGC becomes invariant to nonlinear changes in brightness and contrast. During process variations, the object interiors may exhibit unpredictable brightness changes. However, irrespective of the brightness changes, the object boundaries are always visible. Basically, only the shapes of objects remain the same. Geometry-based finders handle unpredictable brightness variations by operating only on the shape features of a template. In this case, shape is nothing but a linked list of significant edge pixels. Unless the complete geometry of the object to be recognized is acquired from a CAD model, how does a geometry-based finder know the object s exact geometry based on a single image of the object? Quantization of the edges in the image creates subtle variations of lines and curves; shadows and highlights may look like geometric features but are in fact lighting artifacts. Is the shape in the following image (Fig. 4.) a perfect square?
8 Fig. 4. Square or a five-sided polygon? To determine if this is a square or a five-sided polygon, the user must explicitly edit the model or at least specify an error tolerance for fitting lines. Any geometry-based recognition algorithm must use an error tolerance for fitting lines, circular arcs, and other curves. This procedure introduces approximations (and errors) to the model before runtime recognition. Then at run time, as lines and curves are fit to the edges in the new image, additional errors are introduced. NGC on the other hand operates on the image without estimating edge geometry. Errors are not introduced during training or run time. To handle nonlinear changes in contrast, vsfind operates on edge images (e.g., the result of a Sobel filter). This is typically the first step to geometry-based processing, too, because lines and curves are fit to the edges in the image. If a nonlinear contrast transformation is applied to the gray-scale image, it affects the magnitudes of the edges in the edge image, but not the sub-pixel locations of the edges. vsfind fits the grayscale edges in the template to the gray-scale edges in the image for precise recognition and registration. Finally, the cost of geometry-based analysis increases at least linearly with the complexity of the scene. The complexity of the scene can be measured by the number of separate lines and curves in the image or by the number of pixels corresponding to the zero crossings of edges above some threshold. The speed of vsfind, however, is basically independent of scene complexity because it is based on NGC. Handling multilayer buildup in wafers As semiconductor layers are added to a wafer, new features appear in the image. The edges in the image are first thresholded, using either an automatic threshold or a specified threshold. Then isolated edge pixels are filtered and only the significant edge pixels are kept and used in training. This way, only these pixels (most probably the boundary pixels) would be selected and hence any objects that may clutter the interiors at a later time would not have any effect on the correlation score.
9 Handling multiple instances of a template Since an exhaustive search is performed at the top of the data pyramid to locate the trained template and then the correlation scores are sorted, no additional computations (except for the hill climbing) are necessary to find multiple instances of a template. The next better matches are simply pursued through the successive levels of the data pyramid. Handling partly visible, overlapping, touching templates By definition, sparse (point) correlation correlates only on a set of pre-selected points to locate the best match. Hence, it can handle situations wherein parts of the templates are cluttered, overlapping, touching, or not visible due to clipping by the field of view. However, if the portion of the template that is missing happens to be where most of the points were selected during training, then this algorithm does fail. However, the training can be forced to select uniformly distributed points to avoid this failure. Handling templates of arbitrary shapes In sparse correlation, the points that are not selected during training are masked out of the template data during the correlation score computations. Users can also mask out portions of the template that are not important for recognition. In fact, any arbitrarily shaped template can be defined. Performance of the Proposed Pattern-Matching Algorithm: vsfind To test the performance of vsfind, the following target sheet (Fig. 5.) was placed on a micrometer stage and repeatedly moved under the camera. For all tests, the square bulls-eye target in the middle was trained as the template. The template was 191 x 191 pixels in size within a 640 x 480 image. Accuracy tests were performed on both linear and rotating micrometer stages. As the target was moved under the camera, images were acquired and the template was found by vsfind. When the target was moved on the linear stage, vsfind was run with an angular range of ± 0 degrees, so only an (X,Y) translation was sought. When the target was moved on the rotating stage, both the angle and (X,Y) translation were sought.
10 Fig. 5. Target image used in accuracy tests. Twenty pictures each were taken, processed, and stored for the translation test and the rotation test. The accuracy parameter to vsfind (on a scale of 1 to 100) was set to a value of 30 for these tests. The following graph (Fig. 6.) shows the linearity of found locations for the twenty translation images: For these tests, the target was moved almost vertically in steps of about 1/3 pixels. A line was least-squares fit to the twenty found locations. The vsfind: Nonlinearity in Pixels plot below (Fig. 6.) shows the deviation from the fit line for each of the twenty sample images. With this test, questions arose about the accuracy of the micrometer stage and any minute motion of the camera. To answer these questions, the four squares in the corners of the images were used as fiducials. The centers of the four square fiducials were found by applying a high-precision line fitter to each side of each square. By intersecting the four lines around each square, four accurate corners for each square were derived. Then the (X,Y) locations of the four corners of each square were averaged to get the centroid of each square. Finally, the centroids of the four squares were averaged to get a single (X,Y) location representing the centroid of the square fiducials. The linearity of the fiducials travel on the micrometer stage was analyzed in the same way as the linearity of the finder s results was analyzed, above. The graph of non-linearity is in shown in Fig. 7.
11 vsfind: Nonlinearity in Pixels Std. Dev.: 0.016; Max: pixels Pixels Sample Images Fig. 6. vsfind: Nonlinearity in pixels. Fiducials: Nonlinearity in Pixels Std. Dev.: 0.013; Max: pixels Pixels Sample Images Fig. 7. Fiducials: Nonlinearity in pixels. Clearly, the errors from vsfind and the metrology (line-fitter) tools are in correspondence, yet the tools compute the target location using completely different software and methodologies. Consequently, movement of the camera or non-linear movement of the stage caused some error. To determine an absolute error that is independent of these movements, the finder locations were simply subtracted from the combined fiducial centroids in each image. The plot of these relative errors is shown in Fig. 8. For the rotation test, the target was rotated in steps of about 0.42 degrees, so the twenty images covered a total of about 8 degrees of travel. Fig. 9. compares the angular difference in the angle reported by vsfind for the rotated template with the combined rotational angle of the four square fiducials.
12 When solving for rotation and translation instead of translation only, the error in translation increases, as depicted in Fig. 10. Finder Error in Translation Relative to Fiducials Std. Dev.: 0.011; Max: pixels Pixels Sample Images Fig. 8. vsfind s error in translation relative to fiducials. Finder Error in Rotation Relative to Fiducials Std. Dev.: ; Max: degrees Degrees Sample Images Fig. 9. vsfind s error in rotation relative to fiducials.
13 Finder Error in Translation Relative to Fiducials While Also Solving for Rotation Std. Dev.: 0.023; Max: pixels Pixels Sample Images Fig. 10. vsfind s error in translation relative to fiducials while also solving for rotation. All of the above graphs illustrate the results of tests performed with the finder s accuracy set to 30. The following table summarizes comparable results for accuracy settings of 10 and 20. Included are the search times on a 400 MHz Pentium II for an angular search range of ±45 degrees. Accuracy Settings Translation Test Recognition time (ms) One standard deviation (S.D.) error (pixels) Rotation Test Recognition time (ms) One S.D. error in translation (pixels) One S.D. error in rotation (degrees) Summary and Conclusions A pattern-matching technique called vsfind was described. vsfind extracts only the essential information from patterns through a training process (patent-pending) built upon normalized correlation (NGC). By using an intelligent search technique and normalized correlation as the matching criteria, vsfind locates trained patterns almost in
14 real time, while maintaining high accuracy. vsfind was favorably compared to geometry-based finders for accurately locating patterns on wafers for registration. Also described was the adaptation of NGC to meet the scale, rotation, calibrated perspective distortion, focus, contrast reversal, missing features, overlapping, touching and linear and non-linear brightness invariant requirements of the vision industries. References 1. Gleason, S.S., M. A. Hunt, and W. B. Jatko Subpixel measurement of image features based on paraboloid surface fit. SPIE Vol Machine Vision Systems Integration in Industry. 2. Gonzalez, R.C., and R. E. Woods Digital image processing. Addison-Wesley Publishing Company, Inc., New York. 3. Jain, A. K Fundamentals of digital image processing. Prentice-Hall, Inc., Englewood Cliffs, New Jersey. 4. Krattenthaler, W., K. J. Mayer, and M. Zeiller Point correlation: a reducedcost template matching technique. First IEEE International Conference on Image Processing November, Austin, Texas. 5. Rosenfeld, A., and A. C. Kak Digital picture processing. Academic Press, New York. 6. Russ, J.C The image processing handbook. CRC Press Inc., Ann Arbor, Michigan.
MATCHING CRITERIA IN TEMPLATE LOCALIZING COMPARATIVE ANALYSIS OF EXPERIMENTAL RESULTS
MATCHING CRITERIA IN TEMPLATE LOCALIZING COMPARATIVE ANALYSIS OF EXPERIMENTAL RESULTS Yulka PETKOVA*, Dimitar TYANEV* *Technical University of Varna, Department of Computer Sciences and Technologies, 1,
More informationRule-based inspection of Wafer surface
Rule-based inspection of Wafer surface N.G. Shankar Z.W. Zhong Euro Technology Pte Ltd School of Mechanical & Production Engineering Tech Place 1 Nanyang Technological University Singapore 569628 Nanyang
More informationImage Processing Fundamentals. Nicolas Vazquez Principal Software Engineer National Instruments
Image Processing Fundamentals Nicolas Vazquez Principal Software Engineer National Instruments Agenda Objectives and Motivations Enhancing Images Checking for Presence Locating Parts Measuring Features
More informationCh 22 Inspection Technologies
Ch 22 Inspection Technologies Sections: 1. Inspection Metrology 2. Contact vs. Noncontact Inspection Techniques 3. Conventional Measuring and Gaging Techniques 4. Coordinate Measuring Machines 5. Surface
More informationUsing Edge Detection in Machine Vision Gauging Applications
Application Note 125 Using Edge Detection in Machine Vision Gauging Applications John Hanks Introduction This application note introduces common edge-detection software strategies for applications such
More informationProduct information. Hi-Tech Electronics Pte Ltd
Product information Introduction TEMA Motion is the world leading software for advanced motion analysis. Starting with digital image sequences the operator uses TEMA Motion to track objects in images,
More informationOBJECT SORTING IN MANUFACTURING INDUSTRIES USING IMAGE PROCESSING
OBJECT SORTING IN MANUFACTURING INDUSTRIES USING IMAGE PROCESSING Manoj Sabnis 1, Vinita Thakur 2, Rujuta Thorat 2, Gayatri Yeole 2, Chirag Tank 2 1 Assistant Professor, 2 Student, Department of Information
More informationPredictive Interpolation for Registration
Predictive Interpolation for Registration D.G. Bailey Institute of Information Sciences and Technology, Massey University, Private bag 11222, Palmerston North D.G.Bailey@massey.ac.nz Abstract Predictive
More informationOutline 7/2/201011/6/
Outline Pattern recognition in computer vision Background on the development of SIFT SIFT algorithm and some of its variations Computational considerations (SURF) Potential improvement Summary 01 2 Pattern
More informationPictures at an Exhibition
Pictures at an Exhibition Han-I Su Department of Electrical Engineering Stanford University, CA, 94305 Abstract We employ an image identification algorithm for interactive museum guide with pictures taken
More informationBSB663 Image Processing Pinar Duygulu. Slides are adapted from Selim Aksoy
BSB663 Image Processing Pinar Duygulu Slides are adapted from Selim Aksoy Image matching Image matching is a fundamental aspect of many problems in computer vision. Object or scene recognition Solving
More informationChapter 3 Image Registration. Chapter 3 Image Registration
Chapter 3 Image Registration Distributed Algorithms for Introduction (1) Definition: Image Registration Input: 2 images of the same scene but taken from different perspectives Goal: Identify transformation
More informationEE368 Project Report CD Cover Recognition Using Modified SIFT Algorithm
EE368 Project Report CD Cover Recognition Using Modified SIFT Algorithm Group 1: Mina A. Makar Stanford University mamakar@stanford.edu Abstract In this report, we investigate the application of the Scale-Invariant
More informationCIRCULAR MOIRÉ PATTERNS IN 3D COMPUTER VISION APPLICATIONS
CIRCULAR MOIRÉ PATTERNS IN 3D COMPUTER VISION APPLICATIONS Setiawan Hadi Mathematics Department, Universitas Padjadjaran e-mail : shadi@unpad.ac.id Abstract Geometric patterns generated by superimposing
More informationIntroduction.
Product information Image Systems AB Main office: Ågatan 40, SE-582 22 Linköping Phone +46 13 200 100, fax +46 13 200 150 info@imagesystems.se, Introduction TEMA Automotive is the world leading system
More informationAn Automated Image-based Method for Multi-Leaf Collimator Positioning Verification in Intensity Modulated Radiation Therapy
An Automated Image-based Method for Multi-Leaf Collimator Positioning Verification in Intensity Modulated Radiation Therapy Chenyang Xu 1, Siemens Corporate Research, Inc., Princeton, NJ, USA Xiaolei Huang,
More information(Refer Slide Time 00:17) Welcome to the course on Digital Image Processing. (Refer Slide Time 00:22)
Digital Image Processing Prof. P. K. Biswas Department of Electronics and Electrical Communications Engineering Indian Institute of Technology, Kharagpur Module Number 01 Lecture Number 02 Application
More informationAdvanced Vision Guided Robotics. David Bruce Engineering Manager FANUC America Corporation
Advanced Vision Guided Robotics David Bruce Engineering Manager FANUC America Corporation Traditional Vision vs. Vision based Robot Guidance Traditional Machine Vision Determine if a product passes or
More informationDetecting motion by means of 2D and 3D information
Detecting motion by means of 2D and 3D information Federico Tombari Stefano Mattoccia Luigi Di Stefano Fabio Tonelli Department of Electronics Computer Science and Systems (DEIS) Viale Risorgimento 2,
More informationTexture Segmentation by Windowed Projection
Texture Segmentation by Windowed Projection 1, 2 Fan-Chen Tseng, 2 Ching-Chi Hsu, 2 Chiou-Shann Fuh 1 Department of Electronic Engineering National I-Lan Institute of Technology e-mail : fctseng@ccmail.ilantech.edu.tw
More information09/11/2017. Morphological image processing. Morphological image processing. Morphological image processing. Morphological image processing (binary)
Towards image analysis Goal: Describe the contents of an image, distinguishing meaningful information from irrelevant one. Perform suitable transformations of images so as to make explicit particular shape
More informationEE795: Computer Vision and Intelligent Systems
EE795: Computer Vision and Intelligent Systems Spring 2012 TTh 17:30-18:45 FDH 204 Lecture 14 130307 http://www.ee.unlv.edu/~b1morris/ecg795/ 2 Outline Review Stereo Dense Motion Estimation Translational
More informationLocal Image Registration: An Adaptive Filtering Framework
Local Image Registration: An Adaptive Filtering Framework Gulcin Caner a,a.murattekalp a,b, Gaurav Sharma a and Wendi Heinzelman a a Electrical and Computer Engineering Dept.,University of Rochester, Rochester,
More informationStereo Vision Image Processing Strategy for Moving Object Detecting
Stereo Vision Image Processing Strategy for Moving Object Detecting SHIUH-JER HUANG, FU-REN YING Department of Mechanical Engineering National Taiwan University of Science and Technology No. 43, Keelung
More informationSubpixel Corner Detection Using Spatial Moment 1)
Vol.31, No.5 ACTA AUTOMATICA SINICA September, 25 Subpixel Corner Detection Using Spatial Moment 1) WANG She-Yang SONG Shen-Min QIANG Wen-Yi CHEN Xing-Lin (Department of Control Engineering, Harbin Institute
More informationComparison between Various Edge Detection Methods on Satellite Image
Comparison between Various Edge Detection Methods on Satellite Image H.S. Bhadauria 1, Annapurna Singh 2, Anuj Kumar 3 Govind Ballabh Pant Engineering College ( Pauri garhwal),computer Science and Engineering
More informationLocal Feature Detectors
Local Feature Detectors Selim Aksoy Department of Computer Engineering Bilkent University saksoy@cs.bilkent.edu.tr Slides adapted from Cordelia Schmid and David Lowe, CVPR 2003 Tutorial, Matthew Brown,
More informationEE368 Project: Visual Code Marker Detection
EE368 Project: Visual Code Marker Detection Kahye Song Group Number: 42 Email: kahye@stanford.edu Abstract A visual marker detection algorithm has been implemented and tested with twelve training images.
More informationTracking Trajectories of Migrating Birds Around a Skyscraper
Tracking Trajectories of Migrating Birds Around a Skyscraper Brian Crombie Matt Zivney Project Advisors Dr. Huggins Dr. Stewart Abstract In this project, the trajectories of birds are tracked around tall
More informationEE795: Computer Vision and Intelligent Systems
EE795: Computer Vision and Intelligent Systems Spring 2012 TTh 17:30-18:45 WRI C225 Lecture 04 130131 http://www.ee.unlv.edu/~b1morris/ecg795/ 2 Outline Review Histogram Equalization Image Filtering Linear
More informationCPSC 425: Computer Vision
CPSC 425: Computer Vision Image Credit: https://docs.adaptive-vision.com/4.7/studio/machine_vision_guide/templatematching.html Lecture 9: Template Matching (cont.) and Scaled Representations ( unless otherwise
More informationA New Technique of Extraction of Edge Detection Using Digital Image Processing
International OPEN ACCESS Journal Of Modern Engineering Research (IJMER) A New Technique of Extraction of Edge Detection Using Digital Image Processing Balaji S.C.K 1 1, Asst Professor S.V.I.T Abstract:
More informationEECS490: Digital Image Processing. Lecture #19
Lecture #19 Shading and texture analysis using morphology Gray scale reconstruction Basic image segmentation: edges v. regions Point and line locators, edge types and noise Edge operators: LoG, DoG, Canny
More informationTwo ducial marks are used to solve the image rotation []. Tolerance of about 1 is acceptable for NCS due to its excellent matching ability and reliabi
Fast Search Algorithms for IC rinted Mark Quality Inspection 1 Ming-Ching Chang Hsien-Yei Chen y Chiou-Shann Fuh z June 7, 1999 Abstract This paper presents an eective and general purpose search algorithm
More informationSUMMARY: DISTINCTIVE IMAGE FEATURES FROM SCALE- INVARIANT KEYPOINTS
SUMMARY: DISTINCTIVE IMAGE FEATURES FROM SCALE- INVARIANT KEYPOINTS Cognitive Robotics Original: David G. Lowe, 004 Summary: Coen van Leeuwen, s1460919 Abstract: This article presents a method to extract
More informationRange Imaging Through Triangulation. Range Imaging Through Triangulation. Range Imaging Through Triangulation. Range Imaging Through Triangulation
Obviously, this is a very slow process and not suitable for dynamic scenes. To speed things up, we can use a laser that projects a vertical line of light onto the scene. This laser rotates around its vertical
More informationImage Sampling and Quantisation
Image Sampling and Quantisation Introduction to Signal and Image Processing Prof. Dr. Philippe Cattin MIAC, University of Basel 1 of 46 22.02.2016 09:17 Contents Contents 1 Motivation 2 Sampling Introduction
More informationSolving Word Jumbles
Solving Word Jumbles Debabrata Sengupta, Abhishek Sharma Department of Electrical Engineering, Stanford University { dsgupta, abhisheksharma }@stanford.edu Abstract In this report we propose an algorithm
More informationMotion Estimation. There are three main types (or applications) of motion estimation:
Members: D91922016 朱威達 R93922010 林聖凱 R93922044 謝俊瑋 Motion Estimation There are three main types (or applications) of motion estimation: Parametric motion (image alignment) The main idea of parametric motion
More informationImage Sampling & Quantisation
Image Sampling & Quantisation Biomedical Image Analysis Prof. Dr. Philippe Cattin MIAC, University of Basel Contents 1 Motivation 2 Sampling Introduction and Motivation Sampling Example Quantisation Example
More informationMultimedia Technology CHAPTER 4. Video and Animation
CHAPTER 4 Video and Animation - Both video and animation give us a sense of motion. They exploit some properties of human eye s ability of viewing pictures. - Motion video is the element of multimedia
More informationCSE 252B: Computer Vision II
CSE 252B: Computer Vision II Lecturer: Serge Belongie Scribes: Jeremy Pollock and Neil Alldrin LECTURE 14 Robust Feature Matching 14.1. Introduction Last lecture we learned how to find interest points
More informationFeature Detectors - Canny Edge Detector
Feature Detectors - Canny Edge Detector 04/12/2006 07:00 PM Canny Edge Detector Common Names: Canny edge detector Brief Description The Canny operator was designed to be an optimal edge detector (according
More informationComplex Sensors: Cameras, Visual Sensing. The Robotics Primer (Ch. 9) ECE 497: Introduction to Mobile Robotics -Visual Sensors
Complex Sensors: Cameras, Visual Sensing The Robotics Primer (Ch. 9) Bring your laptop and robot everyday DO NOT unplug the network cables from the desktop computers or the walls Tuesday s Quiz is on Visual
More informationDepth. Common Classification Tasks. Example: AlexNet. Another Example: Inception. Another Example: Inception. Depth
Common Classification Tasks Recognition of individual objects/faces Analyze object-specific features (e.g., key points) Train with images from different viewing angles Recognition of object classes Analyze
More informationHUMAN COMPUTER INTERFACE BASED ON HAND TRACKING
Proceedings of MUSME 2011, the International Symposium on Multibody Systems and Mechatronics Valencia, Spain, 25-28 October 2011 HUMAN COMPUTER INTERFACE BASED ON HAND TRACKING Pedro Achanccaray, Cristian
More informationSchedule for Rest of Semester
Schedule for Rest of Semester Date Lecture Topic 11/20 24 Texture 11/27 25 Review of Statistics & Linear Algebra, Eigenvectors 11/29 26 Eigenvector expansions, Pattern Recognition 12/4 27 Cameras & calibration
More informationVisionGauge OnLine Spec Sheet
VisionGauge OnLine Spec Sheet VISIONx INC. www.visionxinc.com Powerful & Easy to Use Intuitive Interface VisionGauge OnLine is a powerful and easy-to-use machine vision software for automated in-process
More informationHistograms of Oriented Gradients
Histograms of Oriented Gradients Carlo Tomasi September 18, 2017 A useful question to ask of an image is whether it contains one or more instances of a certain object: a person, a face, a car, and so forth.
More informationCapturing, Modeling, Rendering 3D Structures
Computer Vision Approach Capturing, Modeling, Rendering 3D Structures Calculate pixel correspondences and extract geometry Not robust Difficult to acquire illumination effects, e.g. specular highlights
More informationEstimation of common groundplane based on co-motion statistics
Estimation of common groundplane based on co-motion statistics Zoltan Szlavik, Laszlo Havasi 2, Tamas Sziranyi Analogical and Neural Computing Laboratory, Computer and Automation Research Institute of
More informationTexture Image Segmentation using FCM
Proceedings of 2012 4th International Conference on Machine Learning and Computing IPCSIT vol. 25 (2012) (2012) IACSIT Press, Singapore Texture Image Segmentation using FCM Kanchan S. Deshmukh + M.G.M
More informationUnderstanding Tracking and StroMotion of Soccer Ball
Understanding Tracking and StroMotion of Soccer Ball Nhat H. Nguyen Master Student 205 Witherspoon Hall Charlotte, NC 28223 704 656 2021 rich.uncc@gmail.com ABSTRACT Soccer requires rapid ball movements.
More informationHISTOGRAMS OF ORIENTATIO N GRADIENTS
HISTOGRAMS OF ORIENTATIO N GRADIENTS Histograms of Orientation Gradients Objective: object recognition Basic idea Local shape information often well described by the distribution of intensity gradients
More informationAvailable online at ScienceDirect. Energy Procedia 69 (2015 )
Available online at www.sciencedirect.com ScienceDirect Energy Procedia 69 (2015 ) 1885 1894 International Conference on Concentrating Solar Power and Chemical Energy Systems, SolarPACES 2014 Heliostat
More informationDigital Makeup Face Generation
Digital Makeup Face Generation Wut Yee Oo Mechanical Engineering Stanford University wutyee@stanford.edu Abstract Make up applications offer photoshop tools to get users inputs in generating a make up
More informationSegmentation and Tracking of Partial Planar Templates
Segmentation and Tracking of Partial Planar Templates Abdelsalam Masoud William Hoff Colorado School of Mines Colorado School of Mines Golden, CO 800 Golden, CO 800 amasoud@mines.edu whoff@mines.edu Abstract
More informationChapter 11 Arc Extraction and Segmentation
Chapter 11 Arc Extraction and Segmentation 11.1 Introduction edge detection: labels each pixel as edge or no edge additional properties of edge: direction, gradient magnitude, contrast edge grouping: edge
More informationCentre for Digital Image Measurement and Analysis, School of Engineering, City University, Northampton Square, London, ECIV OHB
HIGH ACCURACY 3-D MEASUREMENT USING MULTIPLE CAMERA VIEWS T.A. Clarke, T.J. Ellis, & S. Robson. High accuracy measurement of industrially produced objects is becoming increasingly important. The techniques
More informationBlur Space Iterative De-blurring
Blur Space Iterative De-blurring RADU CIPRIAN BILCU 1, MEJDI TRIMECHE 2, SAKARI ALENIUS 3, MARKKU VEHVILAINEN 4 1,2,3,4 Multimedia Technologies Laboratory, Nokia Research Center Visiokatu 1, FIN-33720,
More informationDiscuss Proven technologies that addresses
Robotics and Machine Vision for assembly -Auto Teach, Vision guidance, Color & 3D Mar 5-12 2007 Agenda Discuss Proven technologies that addresses o Quick Tool Bring up o Using Non-touch Vision based Auto
More informationComputer Vision Lecture 17
Computer Vision Lecture 17 Epipolar Geometry & Stereo Basics 13.01.2015 Bastian Leibe RWTH Aachen http://www.vision.rwth-aachen.de leibe@vision.rwth-aachen.de Announcements Seminar in the summer semester
More informationThe Lucas & Kanade Algorithm
The Lucas & Kanade Algorithm Instructor - Simon Lucey 16-423 - Designing Computer Vision Apps Today Registration, Registration, Registration. Linearizing Registration. Lucas & Kanade Algorithm. 3 Biggest
More informationECE 176 Digital Image Processing Handout #14 Pamela Cosman 4/29/05 TEXTURE ANALYSIS
ECE 176 Digital Image Processing Handout #14 Pamela Cosman 4/29/ TEXTURE ANALYSIS Texture analysis is covered very briefly in Gonzalez and Woods, pages 66 671. This handout is intended to supplement that
More informationComputer Vision Lecture 17
Announcements Computer Vision Lecture 17 Epipolar Geometry & Stereo Basics Seminar in the summer semester Current Topics in Computer Vision and Machine Learning Block seminar, presentations in 1 st week
More informationAn Image Based Approach to Compute Object Distance
An Image Based Approach to Compute Object Distance Ashfaqur Rahman * Department of Computer Science, American International University Bangladesh Dhaka 1213, Bangladesh Abdus Salam, Mahfuzul Islam, and
More informationAnno accademico 2006/2007. Davide Migliore
Robotica Anno accademico 6/7 Davide Migliore migliore@elet.polimi.it Today What is a feature? Some useful information The world of features: Detectors Edges detection Corners/Points detection Descriptors?!?!?
More informationAutomatic Hybrid Genetic Algorithm Based Printed Circuit Board Inspection
Automatic Hybrid Genetic Algorithm Based Printed Circuit Board Inspection Syamsiah Mashohor, Jonathan R. Evans and Ahmet T. Erdogan School of Engineering and Electronics University of Edinburgh Edinburgh
More informationFeature Detectors - Sobel Edge Detector
Page 1 of 5 Sobel Edge Detector Common Names: Sobel, also related is Prewitt Gradient Edge Detector Brief Description The Sobel operator performs a 2-D spatial gradient measurement on an image and so emphasizes
More informationDetection of Edges Using Mathematical Morphological Operators
OPEN TRANSACTIONS ON INFORMATION PROCESSING Volume 1, Number 1, MAY 2014 OPEN TRANSACTIONS ON INFORMATION PROCESSING Detection of Edges Using Mathematical Morphological Operators Suman Rani*, Deepti Bansal,
More informationRecall: Derivative of Gaussian Filter. Lecture 7: Correspondence Matching. Observe and Generalize. Observe and Generalize. Observe and Generalize
Recall: Derivative of Gaussian Filter G x I x =di(x,y)/dx Lecture 7: Correspondence Matching Reading: T&V Section 7.2 I(x,y) G y convolve convolve I y =di(x,y)/dy Observe and Generalize Derivative of Gaussian
More informationComparison of Some Motion Detection Methods in cases of Single and Multiple Moving Objects
Comparison of Some Motion Detection Methods in cases of Single and Multiple Moving Objects Shamir Alavi Electrical Engineering National Institute of Technology Silchar Silchar 788010 (Assam), India alavi1223@hotmail.com
More informationSegmentation of Range Data for the Automatic Construction of Models of Articulated Objects
Segmentation of Range Data for the Automatic Construction of Models of Articulated Objects A. P. Ashbrook Department of Artificial Intelligence The University of Edinburgh Edinburgh, Scotland anthonya@dai.ed.ac.uk
More informationPerspective Projection Describes Image Formation Berthold K.P. Horn
Perspective Projection Describes Image Formation Berthold K.P. Horn Wheel Alignment: Camber, Caster, Toe-In, SAI, Camber: angle between axle and horizontal plane. Toe: angle between projection of axle
More informationCS443: Digital Imaging and Multimedia Binary Image Analysis. Spring 2008 Ahmed Elgammal Dept. of Computer Science Rutgers University
CS443: Digital Imaging and Multimedia Binary Image Analysis Spring 2008 Ahmed Elgammal Dept. of Computer Science Rutgers University Outlines A Simple Machine Vision System Image segmentation by thresholding
More informationAdvanced Vision System Integration. David Dechow Staff Engineer, Intelligent Robotics/Machine Vision FANUC America Corporation
Advanced Vision System Integration David Dechow Staff Engineer, Intelligent Robotics/Machine Vision FANUC America Corporation Advanced Vision System Integration INTRODUCTION AND REVIEW Introduction and
More informationDigital Image Processing Fundamentals
Ioannis Pitas Digital Image Processing Fundamentals Chapter 7 Shape Description Answers to the Chapter Questions Thessaloniki 1998 Chapter 7: Shape description 7.1 Introduction 1. Why is invariance to
More informationRKUniversity, India. Key Words Digital image processing, Image enhancement, FPGA, Hardware design languages, Verilog.
Volume 4, Issue 2, February 2014 ISSN: 2277 128X International Journal of Advanced Research in Computer Science and Software Engineering Research Paper Available online at: www.ijarcsse.com Image Enhancement
More informationELEC Dr Reji Mathew Electrical Engineering UNSW
ELEC 4622 Dr Reji Mathew Electrical Engineering UNSW Review of Motion Modelling and Estimation Introduction to Motion Modelling & Estimation Forward Motion Backward Motion Block Motion Estimation Motion
More informationOptical Flow Estimation with CUDA. Mikhail Smirnov
Optical Flow Estimation with CUDA Mikhail Smirnov msmirnov@nvidia.com Document Change History Version Date Responsible Reason for Change Mikhail Smirnov Initial release Abstract Optical flow is the apparent
More informationDrywall state detection in image data for automatic indoor progress monitoring C. Kropp, C. Koch and M. König
Drywall state detection in image data for automatic indoor progress monitoring C. Kropp, C. Koch and M. König Chair for Computing in Engineering, Department of Civil and Environmental Engineering, Ruhr-Universität
More informationHOUGH TRANSFORM CS 6350 C V
HOUGH TRANSFORM CS 6350 C V HOUGH TRANSFORM The problem: Given a set of points in 2-D, find if a sub-set of these points, fall on a LINE. Hough Transform One powerful global method for detecting edges
More information3D Rasterization II COS 426
3D Rasterization II COS 426 3D Rendering Pipeline (for direct illumination) 3D Primitives Modeling Transformation Lighting Viewing Transformation Projection Transformation Clipping Viewport Transformation
More informationCS 534: Computer Vision Texture
CS 534: Computer Vision Texture Spring 2004 Ahmed Elgammal Dept of Computer Science CS 534 Ahmed Elgammal Texture - 1 Outlines Finding templates by convolution What is Texture Co-occurrence matrecis for
More informationError-Diffusion Robust to Mis-Registration in Multi-Pass Printing
Error-Diffusion Robust to Mis-Registration in Multi-Pass Printing Zhigang Fan, Gaurav Sharma, and Shen-ge Wang Xerox Corporation Webster, New York Abstract Error-diffusion and its variants are commonly
More informationExperiments with Edge Detection using One-dimensional Surface Fitting
Experiments with Edge Detection using One-dimensional Surface Fitting Gabor Terei, Jorge Luis Nunes e Silva Brito The Ohio State University, Department of Geodetic Science and Surveying 1958 Neil Avenue,
More informationManipulating the Boundary Mesh
Chapter 7. Manipulating the Boundary Mesh The first step in producing an unstructured grid is to define the shape of the domain boundaries. Using a preprocessor (GAMBIT or a third-party CAD package) you
More informationBiometrics Technology: Image Processing & Pattern Recognition (by Dr. Dickson Tong)
Biometrics Technology: Image Processing & Pattern Recognition (by Dr. Dickson Tong) References: [1] http://homepages.inf.ed.ac.uk/rbf/hipr2/index.htm [2] http://www.cs.wisc.edu/~dyer/cs540/notes/vision.html
More informationEdge detection. Stefano Ferrari. Università degli Studi di Milano Elaborazione delle immagini (Image processing I)
Edge detection Stefano Ferrari Università degli Studi di Milano stefano.ferrari@unimi.it Elaborazione delle immagini (Image processing I) academic year 2011 2012 Image segmentation Several image processing
More informationCS4733 Class Notes, Computer Vision
CS4733 Class Notes, Computer Vision Sources for online computer vision tutorials and demos - http://www.dai.ed.ac.uk/hipr and Computer Vision resources online - http://www.dai.ed.ac.uk/cvonline Vision
More informationHIGH SPEED WEIGHT ESTIMATION BY IMAGE ANALYSIS
Abstract HIGH SPEED WEIGHT ESTIMATION BY IMAGE ANALYSIS D.G. Bailey*, K.A. Mercer*, C. Plaw*, R. Ball**, H. Barraclough** *Institute of Information Sciences and Technology Massey University **Institute
More informationDEFORMABLE MATCHING OF HAND SHAPES FOR USER VERIFICATION. Ani1 K. Jain and Nicolae Duta
DEFORMABLE MATCHING OF HAND SHAPES FOR USER VERIFICATION Ani1 K. Jain and Nicolae Duta Department of Computer Science and Engineering Michigan State University, East Lansing, MI 48824-1026, USA E-mail:
More informationImage Segmentation. 1Jyoti Hazrati, 2Kavita Rawat, 3Khush Batra. Dronacharya College Of Engineering, Farrukhnagar, Haryana, India
Image Segmentation 1Jyoti Hazrati, 2Kavita Rawat, 3Khush Batra Dronacharya College Of Engineering, Farrukhnagar, Haryana, India Dronacharya College Of Engineering, Farrukhnagar, Haryana, India Global Institute
More informationCHAPTER 5 MOTION DETECTION AND ANALYSIS
CHAPTER 5 MOTION DETECTION AND ANALYSIS 5.1. Introduction: Motion processing is gaining an intense attention from the researchers with the progress in motion studies and processing competence. A series
More informationA threshold decision of the object image by using the smart tag
A threshold decision of the object image by using the smart tag Chang-Jun Im, Jin-Young Kim, Kwan Young Joung, Ho-Gil Lee Sensing & Perception Research Group Korea Institute of Industrial Technology (
More informationCHAPTER 6 PERCEPTUAL ORGANIZATION BASED ON TEMPORAL DYNAMICS
CHAPTER 6 PERCEPTUAL ORGANIZATION BASED ON TEMPORAL DYNAMICS This chapter presents a computational model for perceptual organization. A figure-ground segregation network is proposed based on a novel boundary
More informationAssignment 3: Edge Detection
Assignment 3: Edge Detection - EE Affiliate I. INTRODUCTION This assignment looks at different techniques of detecting edges in an image. Edge detection is a fundamental tool in computer vision to analyse
More informationModel Based Perspective Inversion
Model Based Perspective Inversion A. D. Worrall, K. D. Baker & G. D. Sullivan Intelligent Systems Group, Department of Computer Science, University of Reading, RG6 2AX, UK. Anthony.Worrall@reading.ac.uk
More informationImage Segmentation Techniques for Object-Based Coding
Image Techniques for Object-Based Coding Junaid Ahmed, Joseph Bosworth, and Scott T. Acton The Oklahoma Imaging Laboratory School of Electrical and Computer Engineering Oklahoma State University {ajunaid,bosworj,sacton}@okstate.edu
More informationA Survey of Light Source Detection Methods
A Survey of Light Source Detection Methods Nathan Funk University of Alberta Mini-Project for CMPUT 603 November 30, 2003 Abstract This paper provides an overview of the most prominent techniques for light
More information