Combination of edge and color information for robust preprocessing in facial image quality assessment

With the aim of developing an automatic system to verify if facial images meet ISO/IEC 19794-5 standards and ICAO requirements, this paper proposes a robust image processing method to obtain information for quality evaluation step. Specifically, an adaptive background segmentation and a robust facial feature extraction (including eye centers, four lip features and chin) are proposed. In background segmentation, background information is provided after applying an edge-based segmentation. A color-based segmentation is then used to deal with shadows. In order to overcome the influence of head poses and illumination, which are the main factors of unsuccessful eye detection, an improvement of the circular filter-based eye detection is used to locate eye centers. To accurately detect lip features and chin regardless expressions and presences of beards or mustaches, the proposed method fuses edge and color information together. An experiment was performed on a subset of the FERET and GTAV database, and the experimental results demonstrated the accuracy and robustness of the proposed method.