Date(s): May 2012.
Source: 2012 5th International Symposium on Communications, Control and Signal Processing, Rome, Italy, 2012, pp. 1-6, doi: 10.1109/ISCCSP.2012.6217829.
Research Summary: Florence Faces dataset addresses the gap between 2D, appearance-based face recognition techniques and fully 3D approaches, designed to simulate realistic surveillance conditions in a controlled setting. The dataset pairs high-resolution (HR) 3D face scans of each subject with several 2D video sequences captured at varying resolution and zoom level. Each subject is recorded in three progressively less controlled settings, first in HD video under controlled conditions, then indoors with a standard PTZ surveillance camera, and finally outdoors under challenging, unconstrained conditions, with three levels of zoom captured in each sequence. This progression is designed to probe how effectively 3D models can support face recognition as real-world surveillance conditions become less cooperative.
3dMD’s Role: Florence Faces’ high-resolution (HR) 3D facial scans were captured on a 3dMDface system, the same underlying Florence dataset presented in more detail here than in the original 2011 workshop paper.
Article: Florence faces: A dataset supporting 2D/3D face recognition.
Authors: AD Bagdanov, A Del Bimbo, I Masi. MICC-Media Integration and Communication Center, University of Florence, Italy.