基於AdaBoost演算法對人臉影像縮短訓練時間及以人臉辨識性別的方法

A Method to Shorten Training Time for Face Image and Gender Recognition based on AdaBoost Algorithm

陳嘉鴻(Jia-Hong Chen)

指導教授 ： 楊致芳

國立臺北大學/電機資訊學院/電機資訊產業研發碩士專班/碩士(2012年)

摘要

在人臉影像上隱含著許多資訊，如年齡、表情、種族、身分、性別...等，本論文探討的主題為希望透過人臉影像資訊來辨識性別，進而可針對較多男性觀賞者或女性觀賞者，播放合適的廣告來提升廣告效益。除此之外，人臉性別辨識系統還可以應用在許多地方，例如單純將大量人臉影像進行性別分類的工作、有性別限制場所的過濾、安全機制...等等。
本篇論文主旨在探討以人臉辨識性別的問題，我們選擇一個基於學習理論的辨識方法，並將其運用在以人臉特徵辨識性別上。在訓練學習系統方面，我們選擇用AdaBoost演算法[21]作為學習機制的基礎，以Viola和Jones在2003年[14]所發表的人臉辨識系統為出發點，並且結合了可調性邊界(Soft Margin)機制補償了過度學習(overfitting)的問題。在擷取資料特徵的部分，除了實現Freund和Schapire[21]提出實驗方法中所使用的Haar-Like矩形特徵[15]之外，我們多加入了七種自行設計的矩形特徵型態，並且嘗試僅訓練半張人臉特徵資料來大幅縮短挑選弱分類器所需耗費的時間；經由這樣的搭配，發展出一套人臉辨識性別系統。

關鍵字

人臉性別辨識 ； 矩形特徵 ； 分類器

並列摘要

Face Image implies much information, such as age, facial expression, race, identity, gender and more. In this report, we adopt the gender recognition based on face image features. The advertisers and managers enable to play the advertisements/commercials that are suitable for more male audience or female audience for higher advertising effectiveness. Besides, the gender recognition system based on face image can further apply to varied scenarios, like gender classification based on numerous face image samples, the gender recognition and classification for specific occasions, security and surveillance.

The purpose in this report is to examine the problem of the gender recognition system based on face image. We apply recognition learning theories to gender recognition based on face image features. For the training and learning system, we select the AdaBoost algorithm on learning mechanism, based mainly on the face recognition system that has been developed by Viola and Jones in 2003, and further combine the Soft Margin Mechanism to compensate for overfitting problems. In the process of data capture, we not only adopt Haar-Like features that proposed by Freund and Schapire, but also five more self-developed rectangle features. In this thesis, we try to substantially shorten the time period of proceeding and analyzing data based on half of face image feature as well as develop the advanced system.

並列關鍵字

Adaboost ； Soft Margin

參考文獻

[1] H. A. Rowley, S. Baluja and T. Kanade, “Neural network-based face detection,” IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), vol. 20, no. 1, pp. 23-38, Jan. 1998.

[4] A. Hadid, M. Pietikainen and T. Ahonen, “A discriminative feature space for detecting and recognizing faces,” IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), vol. 2, pp. 797-804, July 2004.

[8] G. Shakhnarovich, P. A. Viola and B. Moghaddam, “A Unified Learning Framework for Real Time Face Detection and Classification" Fifth IEEE International Conference on Automatic Face and Gesture Recognition, pp.14-21, May 2002.

[10] B. Wu, H. Ai and C. Huang, “LUT – based Adaboost for gender classification,” International Conference on Audio and Video-Based Biometric Person Authentication (AVBPA), vol. 2688, pp. 104-110, June 2003.

[11] B. WU, H. AI and C. Huang, “Real-time gender classification,” Third International Symposium on Multispectral Image Processing and Pattern Recognition, vol. 5286, pp.498-503, Oct. 2003.