FACIAL FEATURE EXTRACTION USING GRADIENT FEATURES AND MQDF MATCHING

摘要

In this paper, we propose a novel approach for the extraction of facial features from frontal face images. The proposed approach involves 3 main processing stages. In the first stage, a bounding box is placed over the face. In the second stage, candidate positions for the facial features are identified using a set of MQDF-based distance functions. Finally, in the third stage, the candidate feature positions are evaluated using an MQDF-based verification function, and an optimal configuration is chosen. Results on the FERET database indicate that with a proper setting of parameters, the proposed algorithm can accurately extract facial features over many different scales.