Accuracy analysis for facial feature detection

摘要

The accuracy of feature-based human face recognition is inherently dependent upon the correctness of the identification of the facial features. However, the extent and details of this dependence are not known. in this paper, we attempt to develop a basic methodology that can be used to discover how sensitive the recognition process is to inaccuracies in facial feature detection from front-view ID-type images.We apply this methodology to face measurements involving the eyes, mouth, cheeks and chin in order to discover how recognition is affected by measurement inaccuracies. We subsequently judge the quality of our facial feature detection modules by using the recognition sensitivity to individual face measurements. An important point of discussion is the link between measurement and feature inaccuracies. Although this study is meant to be preliminary, mainly to design a framework for the methods, the information gained will help in guiding future improvements in facial feature detection.