Machine Vision Giving Eyes to Robots:  Resources in Technology.  (March 1990).  Technology Teacher, 49: 6, 21-28.
 
 This module introduces machine vision, which can be used for inspection, robot guidance and part sorting. The future for machine vision will include new technology and will bring vision systems closer to the ultimate vision processor, the human eye.
 
 
Chen, Kan and Frank P. Stafford.  The Employment Effects of High-Technology: A Case Study of Machine Vision.  Research Report No. 86-19.  National Commission for Employment Policy (DOL), Washington, D.C.
 
 A case study of machine vision was conducted to identify and analyze the employment effects of high technology in general.  (Machine vision is the automatic acquisition and analysis of an image to obtain desired information for use in controlling an industrial activity, such as the visual sensor system that gives eyes to a robot.)  Machine vision as a new industry has taken off on an exponential rise. The total employment in the machine vision industry has been growing rapidly and will continue to increase.  A large portion of the jobs have been taken by highly trained technical people.  However, as the process becomes more standardized, blue-collar workers with additional training will be able to fill some of the jobs. The United States leads the world in the development of machine vision.  It is expected that this industry may help lessen the number of imports coming into the United States and the amount of labor-intensive manufacturing leaving the country.  The near-term prospects for the machine vision industry require the identification of markets with sufficient volume of applications and a process technology to produce the vision systems at a low enough cost to attract users and generate profits in the industry.
 
 
Baker, Eva L.  et al.  Directly Comparing Computer and Human Performance in Language Understanding and Visual Reasoning.  Paper presented at the Annual Meeting of the American Educational Research Association (New Orleans, LA, April 5-9, 1988).
 
 Evaluation models are being developed for assessing artificial intelligence (AI) systems in terms of similar performance by groups of people.  Natural language understanding and vision systems are the areas of concentration.  In simplest terms, the goal is to norm a given natural language system's performance on a sample of people. The specific program under study is a natural language query system, IRUS. IRUS is designed to serve as a general purpose interface to a broad range of databases and expert systems.  In the vision area, common measures of visual tasks are being analyzed in terms of their appropriateness to the vision system. This is the inverse of the language exploration that began with the tasks and created the measures.  A review by the vision community of approaches they would use to compare human and machine vision will determine if looking for consistent benchmarks is a feasible approach.
 
 
Poggio, Tomaso and David Beymer.  (May 1996).  Learning to see.  IEEE Spectrum, 33(5): 60-65+.

Computational models and neurobiological discoveries suggest that the task of recognizing objects (faces among them) requires learning - a process during  which the connections among the nerve cells in the visual cortex are selectively altered.  Seeing requires the same kind of learning and adaptive behaviors as underlie conscious cognition.  Central to understanding the mind - and to developing intelligent machines - is this matter of learning.  From a computational point of view, the field of vision is one of learning's most complex components.  Within just the last few years, approaches to these problems have begun to merge.  The goal of visual neuroscience is to  comprehend how the human visual system works.  Meanwhile, researchers in computer vision are attempting to build machines that see; but these days, instead of hard-wiring a machine or program to solve a specific visual task, they are working on systems that can be trained to independently inspect and recognize unfamiliar objects.
 
 
Leauby, Bruce A.  (Spring 1998).  Industry.  Pennsylvania Cpa Journal, 69(1): 5.
 
 Banks and other financial institutions are looking for fraud-proof methods of personal identification, particularly at ATM locations.  A company has developed a product that offers a solution to this problem through leading-edge computer vision technologies.
 
 
Guerra, Enrique, Alfonso Manriquez, Daniel Schwartz, and J. Rene Villalobos.
 (October 1997).  Three dimensional automated visual inspection of surface mounted devices.  Computers & Industrial Engineering, 33(1,2): 365-368.
 
 Surface mounted technology has created a demand for machine vision systems that can reliably verify component placement for screening inspection and process improvement.  This paper describes a new 3-D computer vision system being developed at the Center for Electronics Manufacturing of the University of Texas at El Paso for the inspection of printed circuit boards.  A small range finder, utilizing a laser-based triangulation measurement method, is being  employed for the development of an effective high resolution 3-D SMT inspection  system.
 
 
Braggins, Don.  (1995).  A critical look at robot vision.  The Industrial Robot, 22(6): 9-12.
 
 One sort of robot vision is the use of robot-mounted cameras for inspection by vision techniques - in effect the vision is not providing data for the robot controller but passing the eye over something to be inspected in much the same way that a human inspector might examine an assembly from a number of different angles. The barriers to robot vision include: 1.  It has been difficult to communicate data about coordinates to the robot.  2. Computer vision deals with the information present in a 2-dimensional representation of the 3-dimensional world.  The future of vision systems includes robust stereo systems that relies on trigonometry to calculate where things are.
 
 
Vaidyanathan, Shankar and Shivakumar Raman.  (September 1995).  OMNE-vision - Object measurement in a noisy environment using vision.  Computers in Industry, 27(1): 23-32.
 
 Object identification is critical for measurement, inspection, and automatic assembly.  Computer vision systems make use of ideal and ingenuous algorithms for object recognition.  The presence of noise in the environment is usually neglected or treated in a simplified manner.  An algorithm is presented for identifying objects in a noisy environment.
 
 
Phan, Duc Truong and Robert J. Alcock.  (April 1998).  Automated grading and defect detection:  A review.  Forest Products Journal, 48(4): 34-42.
 
 In a plant that manufactures wood products, such as lumber, dimension stock, or veneer sheets, inspection of the products is a necessary part of the production process.  At present, most inspection tasks are carried out manually.  However, because of the high speed with which it is necessary to perform the operation and the stress involved, attempts have been made to automate this grading process.  This paper surveys research in the field of automatic inspection of wood, particularly focusing on computer vision techniques.  The methods are put into an Automated Visual Inspection framework, which is subdivided into commonly used modules for image acquisition, image enhancement, image subdivision, feature extraction, and classification.
 
 
Mehta, Dinesh P and Sartaj Sahni.  (December 1997).  Models, techniques, and algorithms for finding, selecting, and displaying patterns in strings and other discrete objects.  Journal of Systems & Software, 39(3): 201-221.
 
 An attempt is made to demonstrate that pattern discovery in discrete objects is an important intellectual task performed by humans that would benefit from automation.  Everyday examples are presented from software engineering and other areas such as document preparation, molecular biology, computer vision, and computer graphics.  A general methodology is proposed for a system that automates pattern discovery.  This methodology has 3 components: 1.  discovery: find all common patterns in objects, 2.  selection: determine which of these patterns is important by taking into account criteria that are important to a particular application or user, and 3.  display: display the selected patterns.  The methodology was developed by taking into account a number of practical considerations in the areas of computer-human interaction and visualization.  Efficient algorithms for some of the problems that must be solved in order to implement the proposed methodology are outlined.
 
 
Su, Chao-Ton, C Alec Chang, and Fang-Chih Tien.  (November 1995).  Neural networks for precise measurement in computer vision systems.  Computers in Industry.  27(3): 225-236.
 
 Although computer vision systems have been successfully applied to some inspection tasks, they are generally not considered as precise measurement tools due to dimensional distortion and errors.  A procedure is presented to correct these errors for precise measurement.  The first step is to formulate calibration models for image coordinate systems using neural networks.  Then neural networks to model dimensional errors from the initial measurement are structured in a learning stage using standard parts.  Finally, these models are used to correct measurement errors in measurement tasks.  These proposed procedures are implemented as an example.
 
 
Grimson, W.E.L. and J.L. Mundy.  (March 1994).  Computer vision applications.
 Communications of the ACM, 37(3): 44-51.
 
 Computer vision provides a primary method for understanding how to make intelligent decisions about an evironment, on the basis of sensory inputs.  Vision systems only receive measurements of reflected brightness as input.  Image brightness is not generally independent, and the goal of computer vision is to determine sufficient additional constraints to invert brightness into scene parameters.  One class of methods achieves inversion by fixing some scene parameters.  A 2nd class of methods achieves inversion by restricting the problem domain.  A 3rd class achieves inversion by acquiring additional images.  In general, vision methods seek to extract scene parameters, such as surface material type and object shape from image brightnesses, and to use such extracted parameters to match against known-object models to support tasks such as recognition.  While general-purpose vision systems remain an area of active research, considerable progress in limited domains such as those listed has led to a number of areas of successful application.
 
 
Ahuja, Narendra.  (September 1995).  On detection and representation of multiscale low-level image structure.  ACM Computing Surveys, 27(3): 304-306.
 
 The objective of computer vision is interpretation of visual images.  Any data-interpretation task of such magnitude requires models of the data.  Some desirable characteristics of strategies for the detection and representation of low-level perceptual structure or multiscale segmentation, which remains an open problem, are described.  The major issues in successful multiscale image segmentation include: 1.  shape and topology invariance, 2.  photometric scaling, and 3. spatial scaling.  The ultimate objective should be to derive a multiscale segmentation of the image and represent it through a hierarchical (usually tree) structure in which the different image segments, their parameters, and their spatial interrelationships are made explicit.
 
 
Magee, Michael and Steven Seida.  (1995).  An industrial model based computer vision system.  Journal of Manufacturing Systems, 14(3): 169-186.
 
 The significant amount of research devoted to model-based vision has not been widely accepted in industrial environments because of the rapid throughput rates typically imposed by industry and the high costs of specialized hardware and prototyping.  The model-based vision system described performs model-based
 reasoning at real-time rates and with low hardware and prototyping costs.  A set of useful features is extracted from observed models using a library of feature-extraction operators. Results show that real-time operation and rapid prototyping are achievable, particularly where there are at least a few nonambiguous model-feature combinations.
 
 
Yeralan, Sencer and Hung-Chang Pai.  (January 1994).  Attributes for expedient computer vision inspection.  IIE Transactions.  26(1): 60-69.
 
 The visual inspection of parts as they progress through the manufacturing  process is an important task in every industry.  Visual inspection, when performed by humans, is a tedious task and given to error.  This is what makes it a good candidate for automation.  Although computer vision systems have been available for over 30 years, the industrial applications of vision systems have become practical only in the last decade.  Image processing and pattern recognition algorithms used in industrial vision systems are built upon a broad body of knowledge in vision research.  However, the use of computer vision systems in quality control has been limited to replicating the visual inspection tasks as they would be performed by a human operator.  It is held that when computerized inspection is employed, quality control inspection plans suitable for computerized inspection should be employed to assure cost-effectiveness.  A simple gauging inspection task is examined and a quality control plan is proposed.
 
 
Arman, Farshid and J.K. Aggarwal.  (March 1993).  Model-based object recognition in dense-range images - A review.  ACM Computing Surveys, 25(1): 5-43.
 
 The goal in computer vision systems is to analyze data collected from the environment and derive an interpretation to complete a specified task.  Vision system tasks may be divided into data acquisition, low-level processing, representation, model construction, and matching subtasks.  A comprehensive survey is presented of model-based vision systems using dense-range images.  In additions, a comprehensive survey of the recent publications in each subtask pertaining to dense-range image object recognition is presented.
 
 
Torkar, Drago, Rudi Murn, and Dusan Pecek.  (1992).  Reflections on light distribution measurement.  Sensor Review, 12(4): 13-16.
 
 An effort to design a reasonably priced computer vision system for light distribution measurement is described.  The application involves the automobile industry, where reflectors are inspected before they are put into use.  The computer vision system is comprised of: 1.  an IBM PC AT computer with color graphics, 2.  a black-and-white video camera with wide angle object glass, 3.  A black-and-white frame grabber with frame store, 4.  a black-and-white control monitor, and 5.  a classic photometer with probe.  The results show that it is possible to attain satisfactory accuracy for inspection with a low cost and simple computer vision system.  This system represents an improvement on the classic method for light measurement.  Its main advantages are time reduction and the possibility of immediate computer data manipulation.
 
 
Asoudegi, Ehsan.
   (November 1992).  Computerized Dimensional Inspection.  Computers & Industrial Engineering, 23(1-4): 357-360.
 
 Within the next decade, as much as 90% of all industrial visual inspection activities might be performed by computer vision technology.  The possibility of real-time visual geometric measurement is examined.  Several factors are involved in obtaining satisfactory measurement of out-of-roundness for a production part at an acceptable processing speed, including: 1.  having a proper illumination system in order to obtain a high-quality image, 2.  eliminating the effect of noise, such as dust, to avoid false detection, and 3.  finding the accurate boundary points.  Algorithms and methods are discussed regarding these factors.
 
 
Elliott, Peter J, John M. Knapman, and Wolfgang Schlegel.  (1992).  Interactive Image Segmentation for Radiation Treatment Planning.  IBM Systems Journal, 31(4): 620-634.
 
 One of the problems that exists in 3-dimensional radiotherapy today is defining tumor volume, target volume, and organs at risk in 3-dimensional computed tomography or magnetic resonance data sets, a necessary precursor to producing a complete 3-dimensional radiation treatment plan. Computer vision in radiology (COVIRA) is a project sponsored by the Commission of the European Communities under the Advanced Informatics in Medicine research program.  COVIRA uses computers to improve the diagnosis and planning of treatment for cancer patients, especially those with brain tumors.  This work has been carried out in an attempt to solve the problem by replacing time-consuming manual image segmentation by computer-assisted interative image segmentation, in which the computer relieves the user of much of the tedious work and enables the operator to use clinical judgment to achieve the desired result.  This article reviews a state-of-the-art workstation ñ the IBM RISC System/6000.
 
 
Anonymous.  Science and Technology:  The Blind Librarians.  (May 30, 1992).  Economist, 323(7761): 85-86.
 
 Machines are expert at recording and storing images, but cannot make even the most rudimentary sense of them without help.  Tasks that humans do not even recognize as such leave computers flummoxed.  Deflummoxing them would open new markets, and make a great deal of money.  The simple ability to see things and compare them with the way they should be, without any more complex understanding of what they are, can be useful.  Cognex, a software company, writes programs that enable computerized manufacturing tools to check the quality of their work.  Unsurprisingly, the defense industry is interested in these abilities.  The broadest use of computer vision may be in looking through databases, not looking at the world.  Computer and telephone companies hope that once computers can search electronic picture libraries in the flexible way humans do, many new uses for these libraries will emerge.
 
 
Griffin, Paul M.  (1990).  3-D Object Pose Determination Using Computer Vision.  Computers & Industrial Engineering, 19(1-4): 215-218.
 
 The principal goal of robot vision is to increase the flexibility with which a robot can interact with its environment.  For example, it is desirable to give the robot the ability to recognize objects and to determine the position and orientation, or pose, of the object so that it can determine how the object should be picked up.  A methodology is presented for the determination of the object pose for general 3-dimensional objects using computer vision.  Three issues are addressed: 1.  how 3-D information about a scene from 2-D projections is obtained, 2.  what constitutes an appropriate object representation to help reduce the complexity of the pose determination problem, and 3.  how to determine the object pose given the 3-D information and the object representation.  By using an enclosing object, the method reduces the number of parameters that must be solved for the nonlinear least squares estimation.
 
 
Parthenis, K., C. Metaxaki-Kossionides, and B. Dimitriadis.  (March 1990).  An Automatic Computer Vision System for Blood Analysis.  Microprocessing & Microprogramming, 28(1-5): 243-246.

An attempt is made to develop a laboratory tool for the field of hematology.  The proposed system deals with the blood analysis problem.  The system processes black and white blood images taken from a CCD camera through a microscope.  All different categories of cells are recognized, counted, and classified.  The white cells are further treated for their classification into the different classes, according to morphological characteristics of their nuclei.  The final classification is printed out in special format for the physician.  The algorithm has been designed in a modular way so that it is easily transferred to other systems.  Using special hardware, the process time can be reduced dramatically.  The researcher's purpose is the creation of a fully automated autonomous and reliable system that will aid physicians in the diagnosis of blood.