Towards Large-Scale 3D Face Recognition

摘要: 3D face recognition holds great promise in achieving robustness to pose, expressions and occlusions. However, algorithms are still far behind their 2D counterparts due the lack of large-scale datasets. We present a model based algorithm for test its performance by combining two large public datasets faces. propose Fully Convolutional Deep Network (FCDN) initialize our algorithm. Reliable seed points then extracted from each evolving level set curves with single curvature dependent adaptive speed function. establish dense correspondence between faces training matching surface around on template ones target A morphable is fitted probe performed parameters gallery Our achieves state art landmark localization results. Face results combined FRGCv2 Bosphorus show that method effective recognizing query real world variations pose expression, occlusion missing data despite huge gallery. Comparing individual accuracy drops when size increases.