| Image segmentation is it fundamental process in the recognition of patterns in digital images, and is defined as the process by which objects in the image are isolated from other objects and the background. This thesis presents a boundary tracking method which locates the boundaries between objects in a scene represented by a digital image by combining elements of edge-based segmentation and boundary-based segmentation. The method employs a Difference-of-Caussians edge detection operator to identify edges, and a neural network to analyze the resulting patterns to determine edge orientation. Historical basis and biological analogues are discussed. A detailed analysis of an edge detection operator is provided and design and implementation of a neural network capable of determining feature orientation is discussed. Weaknesses of the technique, and suggestions for improvement and potential future research, are included. |
