36 years on the pattern recognition front: Lecture given at ICPR’2000 in Barcelona, Spain on the occasion of receiving the K.S. Fu prize.

https://doi.org/10.1016/S0167-8655(02)00211-8Get rights and content

Abstract

The talk consists of four parts: (1) How I became interested in Pattern Recognition (and a bit of personal history), (2) The early years of pattern recognition (late 60’s and 70’s), (3) Pattern recognition in the present (the last 20 years) and (4) Suggestions for the future and prospects.

Section snippets

How I became interested in pattern recognition

It is appropriate to start the lecture by explaining how I became interested in Pattern Recognition which in turn requires a bit of personal history. In the Fall of 1961 I started graduate studies at the University of California at Berkeley. Before arriving there I had spent a few years working in a power plant in Greece and I had been fascinated by control mechanisms and I was looking forward to pursuing a Ph.D. in control theory.

I was in for a major disappointment. Control theory in the

The early years of pattern recognition (1965–1975)

When I entered the field, statistical pattern recognition was the dominant approach. It was essentially an application of non parametric statistics. However, it suffered from a serious problem. Given that some separating hyper-surfaces have been determined from a training set, we could not say much about a solution to the problem without knowing the probability that new samples could be classified correctly. In other words, successful performance on a testing set is not a guarantee for future

Structural pattern recognition

Structural pattern recognition relies on segmentation (of images, curves, etc.) and the extraction of interesting relationships (possibly syntactic) between the parts found during segmentation. Features are defined either as properties of parts or as properties of relationships between parts. They may also be of syntactic nature, namely if a particular syntactic relation holds between certain parts, the value of a boolean feature may be set to one. The features found in this way may be used as

Suggestions for the future

I would like to offer several suggestions for succeeding in pattern recognition research.

(1) It is important to keep a focus on Engineering. Pattern recognition is engineering because we try to design machines that read documents, count blood cells, inspect parts, etc. We must understand the “physics” of the problem and select from amongst available tools the ones appropriate for the problem. It is futile to look for general mathematical/computational techniques that can solve all problems. In

Acknowledgements

I want to thank Henry Baird for arranging a “dry run” of the lecture at Xerox Parc in June of 2000. Henry’s comments and those of David Goldberg, David Fleet, and other members of the audience were very helpful in revising the lecture.

I also want to thank the National Science for the support of my research for close to 30 years by a program first headed by Norman Caplan and then by Howard Moraff. I am particularly grateful that the support came through without strings attached to it and was

References (7)

  • M.J.A. Berry et al.

    Mastering Data Mining

    (2000)
  • A.R. Damasio

    Descartes’ Error: Emotion, Reason, and the Human Brain

    (1994)
  • Goldberg, D., Richardson, C., 1993. Touch-typing with a stylus. In: Proc. INTERCHI’93, Amsterdam, pp....
There are more references available in the full text version of this article.

Cited by (16)

  • Pattern recognition between science and engineering: A red herring?

    2015, Pattern Recognition Letters
    Citation Excerpt :

    As he put it: “It is important to keep a focus on Engineering. Pattern recognition is engineering because we try to design machines that read documents, count blood cells, inspect parts, etc.” ([58] p. 5). More recently, von Luxburg, Williamson and Guyon took a similar stance and, in an attempt to answer the question whether clustering is “art” or “science,” concluded that “if one wants an abstract label, a better one is engineering, which embraces different ways of knowing [...], recognizes the intrinsic psychological component of many problems, has standardized language and problem descriptions to avoid undue technique-focus and most certainly is focused on solving the end-user problem” ([84] p. 76).

  • Recognition of damaged letters based on mathematical fuzzy logic analysis

    2015, Journal of Applied Logic
    Citation Excerpt :

    The damage is additionally increased due to difficulties with distinguishing the characters from their background. Note that our task challenges the principle of Turing test discussed in Pavlidis [11]: the letters on hot ingot are readable easily by humans but with great difficulties by computer. To cope with the level of possible damage, very robust methods are needed.

  • A hybrid method for robust car plate character recognition

    2005, Engineering Applications of Artificial Intelligence
View all citing articles on Scopus

The word “front” in the title should be read in the same way as in the phrase “western front.”

View full text