基於姿勢感知之深度卷積網路的人臉特徵點偵測

Facial Landmark Detection using Pose-Aware Deep Convolutional Network

黃啟清

指導教授 ： 許秋婷

國立清華大學/電機資訊學院/資訊工程學系/碩士(2014年)

https://doi.org/10.6843/NTHU.2014.00491

摘要

人臉特徵點偵測通常會受到各種環境的影響，例如人臉姿勢的不同以及光影變化。我們觀察到人臉姿勢變化是一項影響人臉特徵點偵測準確率的一大因素。為了解決人臉姿勢對偵測準確率的影響，我們利用深度學習的方式去學習一個良好的迴歸器，並且提出了基於姿勢感知的卷積網路來解決姿勢變化的問題。我們首先提出了一個基於卷積網路的分類器來對人臉影像做姿勢的分類，之後我們提出了兩個卷積網路分別對應人臉的不同姿勢來偵測人臉特徵點。此外，我們利用了輪廓的限制來修改修正層。實驗結果驗證了姿勢感知的偵測器可以比原來的偵測器達到更好的效果。

關鍵字

人臉特徵點偵測 ； 深度學習 ； 卷積神經網路

並列摘要

Facial landmark detection usually suffers from the influence by the change of environment, such as pose variation and illumination. We observe that high pose variation is the one most influence the detection accuracy. To tackle the problem of pose variation, we adopt deep learning approach to learn a good regressor and propose a pose-aware CNN to tackle the pose variation. We first develop CNN classifier to classify facial image according to the pose. Next, we develop two CNN to detect the facial landmarks according to the corresponding pose. In addition, we adjust the refinement level by concluding the shape constraint. Our experimental results show that the pose-aware detector performs better than the original landmark detector.

參考文獻

[1] T. F. Cootes, C. J. Taylor, D. H. Cooper, and J. Graham, ”Active shape models their training and application,” Computer Vision and Image Understanding, vol. 61, no.1, pp.38-59, Jan. 1995.

[3] P. Sauer, T. F. Cootes, and C. J. Taylor, ”Accurate regression procedures for active appearance models,” In Proc. BMVC, 2011.

[4] P. A. Tresadern, P. Sauer, and T. F. Cootes, “Additive updatepredictors in active appearance models,” In Proc. BMVC, 2010.

[5] P. N. Belhumeur, D. W. Jacobs, D. J. Kriegman, and N. Kumar, “Localizing parts of faces using a consensus of exemplars,” In Proc. CVPR, 2011.

[6] L. Gu and T. Kanade, “A generative shape regularization model for robust face alignment,” In Proc. ECCV, 2008.