Date of Submission


Date of Award


Institute Name (Publisher)

Indian Statistical Institute

Document Type

Doctoral Thesis

Degree Name

Doctor of Philosophy

Subject Name

Computer Science


Electronics and Communication Sciences Unit (ECSU-Kolkata)


Pal, Nikhil Ranjan (ECSU-Kolkata; ISI)

Abstract (Summary of the Work)

In this thesis we develop several techniques for performing different pattern recognition tasks. In particular, the pattern recognition tasks considered here are classification and vector quantization. We propose several methods for designing classifiers and address various issues involved in the task. For vector quantization, we develop a method for image compression with superior psychovisual reproduction quality. We also propose a method for fast codebook search in a vector quantizer. We exploit different properties of Self-organizing Map (SOM) network for developing these methods. Along with SOM, we also use fuzzy sets theory and Dempster-Shafer theory of evidence to design classi- fiers with enhanced performance. In the thesis we also report results of some empirical studies on the robustness of SOM with respect to topology preservation under several modification of basic SOM. In the following we provide a brief introduction to the pat- tern recognition problem and its different aspects, including the techniques used in the thesis.Pattern recognition (PR) is the most important trait of cognitive ability, be it of humans or animals. The ability to recognize patterns is central to intelligent behavior. We receive signals from environment through our sensory organs which are processed by the brain to generate suitable responses. The whole process involves extraction of information from the sensory signals, processing it using the information stored in the brain to reach a decision that induces some action. All these information we work with are represented as patterns. We recognize voices, known faces, scenes, written letters and a multitude of other objects in our everyday life. In other words, our very survival hinges on our pattern recognition ability. Even more remarkable is the fact that more often than not we perform these tasks in non-ideal or noisy environments. Not only we use pattern recognition with actual signals from the environment, but also we are capable of doing it at intellectual level. For example, faced with a problem of abstract nature, often we recall a similar problem we have faced earlier or have read about, and get a clue to the solution of the problem in hand.The pattern recognition ability is so natural to us that we almost take it for granted. However, the exact details of the underlying process is still mostly shrouded in mystery and is in itself a vast field of research involving several disciplines like neurobiology, psychology etc. Here we are mainly concerned with automated pattern recognition or the pattern recognition tasks performed by machines. Thus we are faced with the task of teaching a machine to recognize patterns. This is, to say the least, a formidable task. Many experts defined the task from different perspectives. Some of them are as follows:Duda and Hart [88] pattern recognition, a field concerned with machine recognition 2 of meaningful regularities in noisy or complex environments.Pavlidis (273) the word pattern is derived from the same root as the word patron and, in its original use, means something which is set up as a perfect example to be imitated. Thus pattern recognition means the identification of the ideal which a given object is made after.Bezdek [31] pattern recognition is a search for structure in data.Schalkoff (301] Pattern recognition (PR) is the science that concerns the description or classification (recognition) of measurements.


ProQuest Collection ID:

Control Number


Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.


Included in

Mathematics Commons