Predictive sampling of facial expression dynamics driven by a latent action space

被引：5

作者：

Boccignone, Giuseppe ^{[1
]}

Bodini, Matteo ^{[1
]}

Cuculo, Vittorio ^{[1
]}

Grossi, Giuliano ^{[1
]}

机构：

[1] Univ Milan, Dept Comp Sci, Milan, Italy

来源：

2018 14TH INTERNATIONAL CONFERENCE ON SIGNAL IMAGE TECHNOLOGY & INTERNET BASED SYSTEMS (SITIS) | 2018年

关键词：

facial expressions; Bayesian filtering; latent variable models; Gaussian Processes; TRACKING; MODEL; RECOGNITION; SIMULATION;

D O I：

10.1109/SITIS.2018.00031

中图分类号：

TP31 [计算机软件];

学科分类号：

081202 ; 0835 ;

摘要：

We present a probabilistic generative model for tracking by prediction the dynamics of affective spacial expressions in videos. The model relies on Bayesian filter sampling of facial landmarks conditioned on motor action parameter dynamics; namely, trajectories shaped by an autoregressive Gaussian Process Latent Variable state-space. The analysis-by synthesis approach at the heart of the model allows for both inference and generation of affective expressions. Robustness of the method to occlusions and degradation of video quality has been assessed on a publicly available dataset.

引用

页码：143 / 150

页数：8

共 44 条

[31] 3-D MOTION ESTIMATION IN MODEL-BASED FACIAL IMAGE-CODING [J].