Hierarchical neural network system for improving edge detection

Szeto, Anthony Wing Kay (1991) Hierarchical neural network system for improving edge detection. Masters thesis, Memorial University of Newfoundland.

[English] PDF (Migrated (PDF/A Conversion) from original format: (application/pdf)) - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (18MB) [English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. (Original Version)

Abstract

This thesis presents a hierarchical neural network system for improving the edge measurements obtained by an edge operator. The neural network system is designed to adjust the edge measurements based on the information provided by neighbouring edges. The adopted strategy is to analyze the local edge patterns to determine and reinforce edge structures while suppressing unwanted noise and false edges. The hierarchical neural network system is made up of four levels of subnets. The subnet in the first level consists of high-order neural nets to determine the potential adjustment on the element of interest by detecting edge contours according to the selected processes in the neural nets and the input local edge pattern. The second level consists of a cooperative-competitive neural net model to determine the orientation of the strongest edge contour in the local edge pattern. The subnet in the third level consists of two types of neural net models. A high-order neural net ascertains the conditions for adjusting the gradient magnitude and determines the amount of adjustment to the gradient magnitude. A semilinear feedforward net is used to compute the new adjusted gradient magnitude and determines if the element of interest is to be an edge element or a non-edge element. The subnet in level four is a semilinear feedforward net which is used to determine the new orientation for the element of interest. A fast learning algorithm is developed to derive suitable weights for the neural nets to perform efficiently and correctly. Using the hierarchical neural network system for each element in the image, highly parallel processing can be achieved. An iterative approach incorporated into the neural network system has also enabled the application of global analysis in the process of adjusting the edge measurements. As a result, the final edge measurements are more accurate.

Actions (login required)

View Item