Benchmark of a large scale database for facial beauty prediction

Abstract

Due to its indefinite evaluation criterion, facial beauty analysis has shown its challenges in pattern recognition and biometric recognition. Plenty of methods have been designed for solving this problem, but there are still various limitations. Most of the available research results in this field are achieved on a small-scale facial beauty database which is difficult in modeling the structure information for facial beauty. And majority of the presented facial beauty prediction algorithm requires burden landmark or expensive optimization procedure. In this paper, we have established a large scale facial beauty database named LSFBD, which is a new database for facial beauty analysis. The LSFBD database contains 20000 labeled images, which has 10000 unconstrained male subjects and 10000 unconstrained female subjects separately, and all images are verified with well-designed rating with average scores and standard deviation. In this database, manually adding the extreme beauty images make the distribution more reasonable and the result more accurate. CRBM or Eigenfaces results are presented in the paper as benchmark for evaluating the task of facial beauty prediction.