An integrated elastic contour fitting and attribute graph matching model for automatic face coding and recognition

Abstract

Describes an integrated system for automatic face coding and recognition. The proposed system consists of three modules: (1) a contour mask fitting module using an active contour model (ACM); (2) an automatic facial landmark extraction module using a Gabor filter; and 3) an elastic graph matching module using an elastic graph dynamic link model (EGDLM). As opposed to contemporary approaches to face recognition, our proposed model provides a fully automatic facial extraction, feature encoding and matching scheme for face recognition. The ORL face database from AT&T Laboratories was adopted for the experiment. Using this portrait gallery of 40 distinct human faces for model training, 400 facial images under varying pose and expression were taken for invariant tests. The experimental results reported an overall correct recognition rate of over 96%.