University of Alberta Multiscale Boundary Identification for Medical

Segmentation of medical image data plays a crucial role in image processing. Manual tracing of the object boundaries by a specialist is a time-consuming task and is subject to operator variability. Edge information contains important characteristics of image content. Fully automatic segmentation procedures are still far from satisfactory in many real situations. In this thesis, the well-known multiscale wavelet-based techniques have been used for edge detection and investigated to enhance boundary identification of medical images. Directional edge detection and subpixel edge resolution are obtained to improve the edge detection performance. Wavelet modulus maxima chain connection through wavelet decay gives a measurement of scale depth of the edge. An edge quality analysis is carried out after using Lipschitz regularity theory to further identify true edges from noise. Experimental results have shown that the proposed multiscale boundary identification method provides superior performance in medical image segmentation.