Selected quality metrics for digital passport photographs

Facial images play a significant role as biometric identifier. The accurate identification of individuals is nowadays becoming more and more important and can have a big impact on security. The good quality of facial images in passport photographs is essential for accurate identification of individuals. The quality acceptance procedure presently used is based on human visual perception and thus subjective and not standardized. Existing algorithms for measuring image quality are applied for all types of images not focused on the quality determination of passport photographs. However there are few documents existing, defining conformance requirements for the determination of digital passport photographs quality. A major document is named "Biometrics Deployment of Machine Readable Travel Documents", published by the International Civil Aviation Organization (ICAO). This thesis deals with the development of metrics for the automated determination of the quality and grade of acceptance of digital passport photographs without having any reference image available. Based on the above mentioned document of the ICAO, quality conformance sentences and related attributes are abstracted with self-developed methods. About fifty passport photographs haven been taken under strictly controlled conditions to fulfill all requirements given by the above mentioned document. Different kinds of algorithms were implemented to determine values for image attributes and to detect the face features. This ground truth database was the source to "translate" natural language into numeric values to describe how "good quality" is represented by numbers. No priority for the evaluation of attributes was given in the ICAO document. For that reason an international online and on-site survey was developed to explore the opinion of user experts whose work is related to passport photographs. They were asked to evaluate the relevance of different types of attributes related to a passport photograph. Based on that survey, weights for the different types of attributes have been calculated. These weights express the different importances of the attributes for the evaluation process. Three different metrics, expressed by the Photograph-/Image-/Biometric Attributes-Quality Indexes (PAQI, IAQI, BAQI) have been developed to obtain reference values for the quality determination of a passport photograph. Experiments are described to show, that the quality of a selected digital passport photograph can be measured and different attributes, which have an impact on the quality and on the recognition of face features can be identified. Critical issues are discussed and the thesis closes with recommendations given for further research approaches.