以特徵熵引導AdaBoost之特徵選取機制於人臉偵測之應用研究

AdaBoost with Entropy-Directed Feature Selection on NBBP for Face Detection

郭家成(Chia-Cheng Kuo)

指導教授 ： 虞台文

大同大學/電機資訊學院/資訊工程學系所/碩士(2010年)

摘要

AdaBoost學習演算法搭配Haar-like特徵在即時人臉偵測上已達到不錯的成效。儘管如此，其學習階段仍是相當的費時。本論文提出使用所謂的鄰接區塊二元圖樣(NBBP)特徵，並結合這些特徵和特徵熵以增進學習效率。藉由計算NBBP特徵熵，每一回合最佳的弱學習子將以非暴力且有系統的方式來決定。此概念可被運用來建置一即時人臉偵測系統。本論文將以ROC曲線和訓練效率與其他相關方法做比較。由結果可知，NBBP特徵在許多方面皆優於Haar-liked特徵，如前者具備光照不變性並擁有較有效的辨別力。特別是我們所提出基於熵的特徵選擇方法提供了一個較為準確的標準，用來評估所擁有特徵之辨別力。因此，此方法可以做為AdaBoost演算法之仲裁者，藉此有效的篩選弱學習子。

關鍵字

AdaBoost ； 人臉偵測 ； 局部二元圖樣 ； 多區塊二元圖樣 ； 鄰接區塊二元圖樣 ； NBBP特徵熵

並列摘要

AdaBoost learning algorithm had achieved good performance for real-time face detection with Haar-like features. Although the great achievement had been reached by AdaBoost, the learning phase is really time-consuming. This thesis introduces the so-called Neighboring-Block Binary Pattern (NBBP) features, and associates each of them with a feature entropy to improve learning efficiency. By computing the entropies of NBBP features, the best weak-classifier of each iteration can be determined systematically in a non-bruteforce manner. The concept is applied to build a real-time face detection system. Various comparisons with other approaches will be presented in the thesis, including receiver operating characteristic (ROC) and training efficiency. It was shown that the NBBP features are intrinsically superior to Haar-like ones in many perspectives, e.g., the former is illumination invariant, and more efficacious in discrimination power. In particular, the proposed feature selection methodology based on entropy provides a precise criterion to measure the discriminability of available features and, hence, can serve as a 'referee' for AdaBoost algorithm to screen weak classifiers effectively.

並列關鍵字

AdaBoost ； Face detection ； LBP ； MB-LBP ； NBBP ； NBBP entropy

參考文獻

[1] P. Viola and M. Jones, “Robust real-time face detection,” in Computer Vision, 2001. ICCV 2001. Proceedings. Eighth IEEE International Conference on, vol. 2, 2001, pp. 747–747.

[2] R. Lienhart and J. Maydt, “An extended set of Haar-like features for rapid object detection,”in Image Processing. 2002. Proceedings. 2002 International Conference on, vol. 1, 2002, pp. I–900 – I–903.

[4] S. Yan, S. Shan, X. Chen, and W. Gao, “Locally Assembled Binary (LAB) feature with feature-centric cascade for fast and accurate face detection,” in Computer Vision and Pattern Recognition, 2008. CVPR 2008. IEEE Conference on, 2008, pp. 1–7.

[5] T. Ojala, M. Pietikainen, and D. Harwood, “A Comparative Study of Texture Measures with Classification Based on Feature Distributions,” PR, vol. 29, no. 1, pp. 51–59, Jan. 1996.

[9] G. Yang and T. Huang, “Human face detection in a complex background,” PR, vol. 27, no. 1, pp. 53–63, Jan. 1994.