The aspect of quality in pattern classification has recently been explored in the context of biometric identification and authentication systems. The results presented in the literature indicate that incorporating information about quality of the input pattern leads to improved classification performance. The quality itself, however, can be defined in a number of ways, and its role in the various stages of pattern classification is often ambiguous or ad hoc.
In this dissertation a more systematic approach to the incorporation of localized quality metrics into the pattern recognition process is developed for the specific task of fingerprint categorization. Quality is defined not as an intrinsic property of the image, but rather in terms of a set of defects introduced to it. A number of fingerprint images have been examined and the important quality defects have been identified and modeled in a mathematically tractable way. The models are flexible and can be used to generate synthetic images that can facilitate algorithm development and large scale, less time consuming performance testing. The effect of quality defects on various stages of the fingerprint recognition process are examined both analytically and empirically. For these defect models, it is shown that the uncertainty of parameter estimates, i.e. extracted fingerprint features, is the key quantity that can be calculated and propagated forward through the stages of the fingerprint classification process. Modified image processing techniques that explicitly utilize local quality metrics in the extraction of features useful in fingerprint classification, such as ridge orientation flow field, are presented and their performance is investigated.
","abstract_html":"<p>The aspect of quality in pattern classification has recently been explored in the context of biometric identification and authentication systems. The results presented in the literature indicate that incorporating information about quality of the input pattern leads to improved classification performance. The quality itself, however, can be defined in a number of ways, and its role in the various stages of pattern classification is often ambiguous or ad hoc.</p><p>In this dissertation a more systematic approach to the incorporation of localized quality metrics into the pattern recognition process is developed for the specific task of fingerprint categorization. Quality is defined not as an intrinsic property of the image, but rather in terms of a set of defects introduced to it. A number of fingerprint images have been examined and the important quality defects have been identified and modeled in a mathematically tractable way. The models are flexible and can be used to generate synthetic images that can facilitate algorithm development and large scale, less time consuming performance testing. The effect of quality defects on various stages of the fingerprint recognition process are examined both analytically and empirically. For these defect models, it is shown that the uncertainty of parameter estimates, i.e. extracted fingerprint features, is the key quantity that can be calculated and propagated forward through the stages of the fingerprint classification process. Modified image processing techniques that explicitly utilize local quality metrics in the extraction of features useful in fingerprint classification, such as ridge orientation flow field, are presented and their performance is investigated.</p>","abstract_has_math":false,"creators":["Mocny, Jakub"],"institution":null,"degree_name":"Doctor of Philosophy","degree_level":"Dissertation","degree_discipline":null,"degree_department":null,"school":null,"contributors":["Michael Carter"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2007,"date_issued":"2007-01-01T08:00:00Z","date_published":"2007-01-01T08:00:00Z","updated_at":"2026-07-24T05:22:25Z","subjects":["Artificial Intelligence","Computer Science"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://scholars.unh.edu/dissertation/412","outbound_label":"Repository record","outbound_source":"dc:identifier"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Michael Carter"]},{"key":"dc:creator","label":"Author","values":["Mocny, Jakub"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"thesis:degree_level","label":"Degree Level","values":["Dissertation"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Doctor of Philosophy"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Artificial Intelligence","Computer Science"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://scholars.unh.edu/dissertation/412"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["The aspect of quality in pattern classification has recently been explored in the context of biometric identification and authentication systems. The results presented in the literature indicate that incorporating information about quality of the input pattern leads to improved classification performance. The quality itself, however, can be defined in a number of ways, and its role in the various stages of pattern classification is often ambiguous or ad hoc.
In this dissertation a more systematic approach to the incorporation of localized quality metrics into the pattern recognition process is developed for the specific task of fingerprint categorization. Quality is defined not as an intrinsic property of the image, but rather in terms of a set of defects introduced to it. A number of fingerprint images have been examined and the important quality defects have been identified and modeled in a mathematically tractable way. The models are flexible and can be used to generate synthetic images that can facilitate algorithm development and large scale, less time consuming performance testing. The effect of quality defects on various stages of the fingerprint recognition process are examined both analytically and empirically. For these defect models, it is shown that the uncertainty of parameter estimates, i.e. extracted fingerprint features, is the key quantity that can be calculated and propagated forward through the stages of the fingerprint classification process. Modified image processing techniques that explicitly utilize local quality metrics in the extraction of features useful in fingerprint classification, such as ridge orientation flow field, are presented and their performance is investigated.
"]},{"key":"dc:title","label":"Title","values":["Error propagation in pattern recognition systems: Impact of quality on fingerprint categorization"]}]}],"canonical_facts":{"dc:contributor":["Michael Carter"],"dc:creator":["Mocny, Jakub"],"dc:description.abstract":["The aspect of quality in pattern classification has recently been explored in the context of biometric identification and authentication systems. The results presented in the literature indicate that incorporating information about quality of the input pattern leads to improved classification performance. The quality itself, however, can be defined in a number of ways, and its role in the various stages of pattern classification is often ambiguous or ad hoc.
In this dissertation a more systematic approach to the incorporation of localized quality metrics into the pattern recognition process is developed for the specific task of fingerprint categorization. Quality is defined not as an intrinsic property of the image, but rather in terms of a set of defects introduced to it. A number of fingerprint images have been examined and the important quality defects have been identified and modeled in a mathematically tractable way. The models are flexible and can be used to generate synthetic images that can facilitate algorithm development and large scale, less time consuming performance testing. The effect of quality defects on various stages of the fingerprint recognition process are examined both analytically and empirically. For these defect models, it is shown that the uncertainty of parameter estimates, i.e. extracted fingerprint features, is the key quantity that can be calculated and propagated forward through the stages of the fingerprint classification process. Modified image processing techniques that explicitly utilize local quality metrics in the extraction of features useful in fingerprint classification, such as ridge orientation flow field, are presented and their performance is investigated.
"],"dc:identifier":["https://scholars.unh.edu/dissertation/412"],"dc:subject":["Artificial Intelligence","Computer Science"],"dc:title":["Error propagation in pattern recognition systems: Impact of quality on fingerprint categorization"],"thesis:degree_level":["Dissertation"],"thesis:degree_name":["Doctor of Philosophy"]},"updated_at":"2026-07-24T05:22:25Z"}