19.4.1 Applications
Machine vision systems have uses in a wide variety of disciplines, from medicine to robotics, from automatic inspection to autonomous navigation, and from document analysis to multimedia systems. New machine vision systems are constantly emerging and becoming a part of everyday life. In this section, we present a brief description some of the various applications of machine vision.
Optical Character Recognition (OCR) and Document Image Analysis
The objective of document image analysis is to recognize the text and graphics components in images and extract the intended information as a human would. Two categories of document processing can be defined, textual processing and graphics processing. Textual processing deals with the text components of a document image. Some tasks here are recognizing the text by optical character recognition (OCR); skew detection and correction (any tilt at which the documents may have been scanned); and finding the columns, paragraphs, text lines, and words. Graphics processing deals with the nontextual components of a document image, such as lines and symbols that make up line diagrams, delimiting lines between text sections, and company logos.
Document analysis is currently a very active area of research. Researchers have devised systems ranging from automatic engineering drawing interpretation and recognition of tabular drawings to the recognition of zip codes from postal envelopes and the interpretation of musical scores. However, the real success story of document analysis is OCR. This is the one area of machine vision in which scientific study has lead to numerous low-cost marketed products. Many of the current OCR systems report accuracies well in the upper 90th percentile.
Many document analysis publications can be found in the journals and conference proceedings listed in the Further Information section. In addition, the International Conference on Document Analysis and Recognition (ICDAR) and the International Workshop or Graphics Recognition (IWGR) are biannual meetings held in conjunction with each other dedicated entirely to the field of document analysis.
Medical Image Analysis
Medical imaging analysis deals primarily with images such as X-rays, computerized tomograph (CT) scans, and magnetic resonance imaging (MRI) images. Early work in medical image analysis overlapped with that of image processing; the main task was to enhance medical images for viewing by a physician; no automatic interpretation or high-level reasoning by the system was involved. However, more recent work is being conducted on medical imagery, which more closely fits the definition of machine vision; for example, systems to search images for diseased organs or tissues based on known models (images) and features of the diseased samples and systems to generate three-dimensional models of organs based on CT scan and MRI. Some active areas of research include the use of three-dimensional images in surgery and surgical planning (especially neurosurgery), virtual surgery, and generating time-varying three-dimensional models obtained from sequences of images (e.g., pumping heart). These systems en- compass many of the aspects of typical machine vision systems plus incorporate many aspects of computer graphics.
In addition to the listings at the end of this section, the following are excellent sources of information on current research in medical imaging: IEEE Transactions on Medical Imaging, IEEE Transactions on Biomedical Engineering, IEEE Transactions on Image Processing, and IEEE Engineering in Medicine and Biology Magazine. There are also a number of conferences dedicated to medical imaging research: SPIE Medical Imaging, IEEE Engineering in Medicine and Biology, Medicine Meets Virtual Reality, select sessions/symposium of IEEE Visualization, and SPIE Visualization in Biomedical Computing.
Photogrammetry and Aerial Image Analysis
Photogrammetry deals with the task of making reliable measurements from images. In the early days of photogrammetry, the images were actual printed photographs often taken from balloons. Today, however, the remote sensing process is multispectral using energy in many other parts of the electromagnetic spectrum, such as ultraviolet and infrared. The images are often transmitted directly from satellites orbiting the globe, such as the Landsat satellites first launched in the early 1970s. Some of the applications of photogrammetry and aerial image analysis include atmospheric exploration; thermal image analysis for energy conservation; monitoring natural resources, crop conditions, land cover, and land use; weather prediction; pollution studies; urban planning; military reconnaissance; plus many others in geology, hydrology, and oceanography.
There are several organizations dedicated to the study of these types of remote sensing tasks. The following are very good sources for information on photogrammetry and aerial image analysis: The American Society of Photogrammetry, The International Remote Sensing Institute (ISRI), and The American Congress on Surveying and Mapping. In addition, there are several conferences and symposia dealing with this topic: SPIE Remote Sensing, International Geosciences and Remote Sensing Symposium, IEEE Transactions on Geoscience and Remote Sensing, and the Symposium on Remote Sensing sponsored by the American Geological Institute.
Industrial Inspection and Robotics
Unlike many of the already mentioned machine vision tasks, automatic inspection and robotics perform many tasks in real time that significantly increases the complexity of the system. The general goal of such systems is sensor-guided control. Typical industrial applications include automatic inspection of machined parts, solder joint surfaces (welds), silicon wafers, produce, and even candy. Some of the challenges
faced by industrial vision system designers include determining the optimal configuration of the camera and lighting, determining the most suitable color space representation of the illumination, modeling various surface reflectance mechanisms, dynamic sensor feedback, real-time manipulator control, real- time operating system interfaces, and neural networks.
Structured lighting techniques have been used extensively in industrial vision applications where the illumination of the scene can be easily controlled. In a typical application, objects on a conveyor belt pass through a plane of light, creating a distortion in the image of the light stripe. The profile of the object at the plane of the light beam is then calculated. This process is repeated at regular intervals as the object moves along the conveyor belt to recover the shape of the object. Then appropriate actions are taken depending on the goal of the system.
The following are sources dedicated primarily to industrial vision applications and robotics: International Journal of Robotics Research, IEEE Transactions on Robotics and Automation, IEEE’s International Conference on Robotics and Automation, and IEEE Transactions on Systems, Man, and Cybernetics.
Autonomous Navigation
Closely related to robotics is the area of autonomous navigation. Much work is being done to develop systems to enable robots or other mobile vehicles to automatically navigate through a specific environment. Techniques involved include active vision (sensor control), neural networks, high-speed stereo vision, three-dimensional vision (range imaging), high-level reasoning for navigational planning, and signal compression and transmission for accurate remote vehicle control.
Visual Information Management Systems
Probably one of the newest areas of machine vision research is in visual information management systems (VIMS). With the recent advances in low-cost computing power and the ever increasing number of multimedia applications, digital imagery is becoming a larger and larger part of every day life. Research in VIMS is providing methods to handle all of this digital information. Such VIMS applications include interactive television, video teleconferencing, digital libraries, video-on-demand, and large-scale video databases. Image processing and machine vision play a very important role in much of this work ranging anywhere from designing video compression schemes, which allow many processing techniques to be performed directly on the compressed datastream and developing more efficient indexing methods for multidimensional data, to automatic scene cut detection for automatically indexing large stockpiles of video data and developing methods to query image databases by image content as opposed to standard structured grey language (SQL) techniques.
Defining Terms
Correspondence problem: The problem of matching points in one image with their corresponding points in a second image.
Histogram: A plot of the frequency of occurrence of the grey levels in an image.
Quantization: The process of representing the continuous range of image intensities by a limited number of discrete grey values.
Sampling: The process of representing the continuous image intensity function as a discrete two-dimensional array.
Segmentation: The process of separating the objects from the background.
Polygonalization: A method of representing a contour by a set of connected straight line segments; for closed curves, these segments form a polygon.
Projection: The transformation and representation of a high-dimensional space in a lesser number of dimensions (i.e., a three-dimensional scene represented as a two-dimensional image).
Thresholding: A method of separating objects from the background by selecting an interval (usually in pixel intensity) and setting any points within the interval to 1 and points outside the interval to 0.
References
Besl, P.J. 1988. Active, Optical range imaging sensors. Machine Vision and Applications 1(2):127–152. Besl, P. and Jain, R.C. 1985. Three dimensional object recognition. ACM Computing Surveys 17(1). Duda, R.O. and Hart, P.E. 1973. Pattern Classification and Scene Analysis. Wiley, New York.
Foley, J.D., van Dam, A., Feiner, S.K., and Hughes, J.F. 1990. Computer Graphics, Principles and Practice.
Addison-Wesley, Reading, MA.
Gonzalez, R.C. and Woods, R.E. 1992. Digital Image Processing. Addison-Wesley, Reading, MA.
Javis, R.A. 1983. A perspective on range finding techniques for computer vision. TEEE Trans. on Pattern
Analysis and Machine Intelligence 5(2):122–139.
Kasturi, R. and Jain, R.C. 1991. Computer Vision: Principles. IEEE Computer Society Press.
Kosko, B. 1992. Neural Networks and Fuzzy Systems. Prentice-Hall, Englewood Cliffs, NJ.
Jain, R., Kasturi, R., and Schunck, B.G. 1995. Machine Vision. McGraw-Hill, New York.
Marr, D. 1982. Vision: A Computational Investigation into the Human Representation and Processing of Visual Information. W. H. Freeman, San Francisco, CA.
Tanimoto, S.L. 1995. The Elements of Artificial Intelligence Using Common Lisp. Computer Science Press.
Winston, P.H. 1992. Artificial Intelligence. Addison-Wesley, Reading, MA.
Further Information
Much of the material presented in this section has been adapted from:
Jain, R., Kasturi, R., and Schunck, B.G. 1995. Machine Vision. McGraw-Hill, New York.
Kasturi, R. and Jain, R.C. 1991. Computer Vision: Principles. IEEE Computer Society Press, Washington.
In addition the following books are reccommended:
Rosenfeld, A. and Kak, A.C. 1982. Digital Picture Processing. Academic Press, Englewood Cliffs, NJ. Jain, A.K. 1989. Fundamentals of Digital Image Processing. Prentice-Hall, New York.
Haralick, R.M. and Shapiro, L.G. 1992–1993. Computer and Robot Vision. Addison-Wesley, Reading, MA. Horn, B. 1986. Robot Vision. McGraw-Hill.
Schalkoff, R.J. 1989. Digital Image Processing and Computer Vision. Wiley, New York. Additional information on machine vision can be found in the following technical journals:
Artificial Intelligence (AI)
Computer Vision, Graphics, and Image Processing (CVGIP)
IEEE Transactions on Image Processing
IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI)
Image and Vision Computing
International Journal of Computer Vision
International Journal on Pattern Recognition and Artificial Intelligence
Machine Vision and Applications (MVA)
Pattern Recognition (PR)
Pattern Recognition Letters (PRL)
The following conference proceedings are also a good source for machine vision information: A series of conferences by the International Society for Optical Engineering (SPIE)
A series of workshops by the Institute of Electrical and Electronic Engineering (IEEE) Computer Society
A series of workshops by the International Association for Pattern Recognition (IAPR)
European Conference on Computer Vision (ECCV)
IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
International Conference on Computer Vision (ICCV)
International Conference on Pattern Recognition (ICPR)