Reconstruction and identification of features in planar and three-dimensional objects

With the increasing complexity of manufacturing techniques, the nondestructive evaluation of products becomes an important component in maintaining product quality and increasing structural reliability. One of the promising new technologies is computerized tomography, which is studied in this dissertation. An important problem in industrial tomography, addressed here, is tomographic reconstruction from incomplete data. Another critical aspect arises from the fact that modern sophisticated nondestructive evaluation techniques, including tomography, generate large amounts of data. It is therefore important to develop computerized methods to analyze this data. One such method that will be investigated in this work is computer-automated feature extraction from images.; Radiography is commonly used for flaw detection in a variety of objects. However, radiographs can be difficult to interpret, especially if the object has a complicated geometry or internal structure. Tomographic reconstructions allow one to see into an object by producing a cross-sectional view of the object density. Conventional tomography requires radiographs from all directions surrounding the object, but in industrial applications body geometry, surrounding structures, or other restrictions can make such complete data collection impossible. Reconstruction from a restricted angle data set is known as the limited angle observation problem. In this dissertation we study analytic continuation and the use of a priori information--of the type available in industrial settings--to perform reconstructions from limited angle data.; Volume reconstructions from computerized tomography or multidirectional radiography produce enormous amounts of data, which raises the additional problem of automated defect recognition. Similar problems arise in real-time automated optical and radiographic inspection systems, where automated inspection provides both reliability and speed. In order to manage the vast amounts of data contained in a raw image, it is useful to reduce the image to a smaller set of important features, and perform subsequent processing on this reduced set. This is a general technique used in many areas of computer vision. In this work we develop feature extraction and comparison methods and demonstrate their applicability to trace break/join defects in printed circuit boards and the related problem of fingerprint identification.