Semi-Automatic Geometric Normalization of Profile Faces using Random Sample Consensus Feature Matching

No ratings

Presented at EISIC 2019 by

Justin Bourlai

This paper proposes a correlation point matching approach, i.e. an efficient methodology for applying geometricnormalization for profile face images. This method is used to increase accuracy without imposing a significantincrease in face matching computational time when using different feature descriptors. In our work, several suchdescriptors are tested to compare the accuracy with which low level facial features (edges), useful for profile faceimage geometric normalization, are extracted. Hence, we determined the most efficient normalization approachthat does not substantially increase computational time. Experimental results show that the use of eigenvaluesproduces a higher than average edge point count, while having a lower increase in computational complexitycompared to other similar algorithms. Then, the extracted features are matched using the random sample consensusalgorithm (RANSAC). Next, the rotational angles between the pairs of features are calculated and averaged toyield the angle of rotation necessary to achieve a proper profile face image normalization representation. Afterapplying our proposed approach to a deep learning-based profile face recognition algorithm, an increase of 7.2%accuracy is achieved when compared to the baseline (non-normalized profile faces). To the best of our knowledge,this is the first time in the open literature that the impact of automated profile face normalization is beinginvestigated to improve deep learning-based profile face matching performance.