Occluded Object Recognition by Neuro-Fuzzy Approach.

Date of Submission

December 1997

Date of Award

Winter 12-12-1998

Institute Name (Publisher)

Indian Statistical Institute

Document Type

Master's Dissertation

Degree Name

Master of Technology

Subject Name

Computer Science


Electronics and Communication Sciences Unit (ECSU-Kolkata)


Ray, Kumar Sankar (ECSU-Kolkata; ISI)

Abstract (Summary of the Work)

1.1 NEURAL NETWORK AN OVERVIEWМОTIVATIONSince from the time the first primitive computing machines are invented, their designers and users are making efforts to push computers beyond the role of automatic calculators into the realm of "thinking machines" That, what is meant by thinking machine is debat- able. The popular phrase "artificial intelligence" is given a variety of definitions. Hence methods for supposedly implementing humanlike thought process with a deterministic machine are varied. Neural networks represent one of these approaches.BIOLOGICAL NEURAL NETWORKSA single cell capable of a sort of crude computation is called a biological neuron. One or more inputs stimulate it and it generates an output that is sent to other neurons. The output in dependent on the strength each of the inputs and on the nature of the input connection ( called a synapse ). Some synapses may be such that an input there will tend to excite the neuron ( increase the output ).Rest may be inhibitory an input to them will tend to reduce the neuron's output. The actual relationship between the input and output is quite complex. Significant time delays may occur between the application of input stimulus and the generation of output response. Fatigue can set in, so that neuron does not always respond in the same way to same inputs. Even random events have an effect on the operation of the neuron. Fotunately, a large body of research indicates that simple models which account for only the most basic neural processes, can provide excellent solutions to practical problemsNEURAL NETWORK CAPABILITIES-ClassificationNeural networks can be used to determine the crop types from satellite photographs, to distinguish a submarine from a boulder given its sonar return, and to identify the diseases of heart from electro cardiograms Any task that can be done by traditional discriminant analysis can be done atleant as well ( and almost always much better ) by neural network.-Noise ReductionAn artificial neural network can be trained to recognise a number of objects. These objects may parts of time-series, images etc. If a version of one of these objectas, corrupted by noise , is presented to a properly trained neural network, the network can provide the original object on which it was trained. This technique has been used with great success in some image restoration problems.-PredictionA very common problem is that of predicting the value of a variable given the historic values of itself( and perhaps of other variables). It has been shown that the neural network, frequently outperform the traditional techniques like ARIMA and frequency domain analysis.Artificial neural networks perform superior to other methods under the following conditions:1. The data on which the conclusions are to be based is "fuzzy". If the input data is human opinions, ill defined categories or is subject to possibly large error then it is better to use artificial neural networks.


ProQuest Collection ID: http://gateway.proquest.com/openurl?url_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:dissertation&res_dat=xri:pqm&rft_dat=xri:pqdiss:28843438

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.



This document is currently not available here.