Optoelectronic digital processors for mathematical morphology and medical image segmentation

The purpose of this dissertation is to present novel optical architectures and associated algorithms for the implementation of morphological image processing operations. These techniques are applied to medical image analysis in order to extract brain matter from human head images acquired through magnetic resonance imaging (MRI).; Mathematical morphology plays an important role in both low and high level image processing. The implementations presented in this work utilize the high fan factor potential of optical systems in order to realize complex morphological operations. The core of the processing elements in these implementations is the optical fiber programmable logic array (OPLA). Compared to purely electronic logic arrays the optoelectronic arrays have the advantage of realizing complex logic functions requiring high fan factors using a minimum number of levels. The OPLAs can operate at clock speeds in the gigahertz range. In this dissertation, an OPLA-based pipeline architecture for morphological image processing is presented. The mapping procedure of the basic mathematical morphology operations to the OPLAs is discussed. The dissertation also presents an iterative parallel algorithm for the watershed transform which is a powerful morphological tool that is often used for image segmentation. The algorithm is tailored to optoelectronic array processors that use OPLAs as processing elements.; One of the current challenges in medical image analysis is the segmentation of MRI data. In the present work, a methodology that segments MR head images in order to extract brain matter is detailed. The technique is based on digital optoelectronic watershed implementation. In order to avoid the oversegmentation problem, frequently encountered with the watershed transform, marker images are used. An algorithm for the selection of these marker images is presented.