Analysis of human attractiveness using manifold kernel regression

This paper uses a recently introduced manifold kernel regression technique to explore the relationship between facial shape and attractiveness on a heterogeneous dataset of over three thousand images gathered from the Web. Using the concept of the Frechet mean of images under a diffeomorphic transformation model, we evolve the average face as a function of attractiveness ratings. Examining these averages and associated deformation maps enables us to discern aggregate shape change trends for male and female faces.