Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


ucin1096665126.pdf (6.35 MB)

ETD Abstract Container

Abstract Header

Proximity Metrics for Contextual Pattern Recognition

Tembe, Waibhav D.
http://rave.ohiolink.edu/etdc/view?acc_num=ucin1096665126

Abstract Details

2004, PhD, University of Cincinnati, Engineering : Computer Science and Engineering.
The principal goal of pattern recognition is to search for interesting associations in observed data and model them as knowledge bases for reliable decision making. At the heart of this problem lies the issue of quantifying proximity between different data and, therefore, investigation of novel ways of quantifying proximity has always been an active area of research. Most popular proximity measures are based on the concept of distance. However, as proved by research in psychology and cognitive science, they fail to model human perception. Surprisingly, little research has been done to introduce the perceptual nature of proximity into pattern recognition systems. This thesis extends the existing research efforts in pattern recognition one step further towards intelligent pattern recognition – pattern recognition that is more compatible to human thinking and perception of patterns. Based on the premise that context plays an important role in perceptual proximity assessment, this thesis proposes new direction in proximity measurement that is not limited by the properties of distance like measures. From this point of view, new concepts for context based assessment of proximity are introduced, mathematically formulated, theoretically analyzed, and experimentally illustrated. The behavior of new formulae for proximity measurement and limitations of classical techniques are discussed using several examples and comparisons. Simultaneously, mathematical relations between context-dependent measures and their classical counterparts have been derived to provide more insight into their properties and highlight their superiority. Further, when integrated into typical unsupervised and supervised pattern recognition algorithms for real world problems, context dependent proximity measures yield promising results. A detailed analysis of the results and comparisons with those obtained using competing classical techniques lead to the conclusion that the context dependent metrics perform better in almost all cases. Possible extensions of context-based pattern recognition are rich with diverse applications. To this end, the thesis suggests several interesting directions for future research in cluster analysis, information theory, analytical geometry, shape recognition and includes an extensive bibliography of relevant research in pattern recognition, machine learning, artificial intelligence, and psychology.
Dr. Anca Ralescu (Advisor)
214 p.
pattern recognition; context-dependent proximity; clustering

Recommended Citations

Citations

  • Tembe, W. D. (2004). Proximity Metrics for Contextual Pattern Recognition [Doctoral dissertation, University of Cincinnati]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1096665126

    APA Style (7th edition)

  • Tembe, Waibhav. Proximity Metrics for Contextual Pattern Recognition. 2004. University of Cincinnati, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=ucin1096665126.

    MLA Style (8th edition)

  • Tembe, Waibhav. "Proximity Metrics for Contextual Pattern Recognition." Doctoral dissertation, University of Cincinnati, 2004. http://rave.ohiolink.edu/etdc/view?acc_num=ucin1096665126

    Chicago Manual of Style (17th edition)

ucin1096665126
707
© 2004, all rights reserved.
This open access ETD is published by University of Cincinnati and OhioLINK.