This paper proposes a novel facial image transformation that minimizes the variation due to aging in facial features. It transforms a face image of an individual according to the probe image to register the facial features. The images are globally registered using Delaunay triangulation which reduces the linear variation in individual's appearance due to aging. Finally, weighted Local Binary Patterns are used to calculate the similarity between the registered images. The proposed approach has been tested on the publicly available FG-NET database [2] and a self-created BROWNS database. It is found to be robust to aging variation to a good extent. Gain in Rank-1 accuracies of 18.36 % and 17.37 % have been achieved on FG-NET and BROWNS database respectively.
展开▼
