使用EMSR法精進PCA之人臉辨識

Improving Face Recognition Performance By EMSR Based on PCA Algorithm

許秀貞(Hsiu-Chen Hsu)

指導教授 ： 駱榮欽

國立臺北科技大學/電資學院/電腦與通訊研究所/碩士(2011年)

https://doi.org/10.6841/NTUT.2011.00321

摘要

主元件分析法(Principal Component Analysis; PCA)是一種可適用於未來智慧家庭的各種設施辨識家中成員，並讓智慧設施依家中成員的個人喜好提供不同貼心服務之即時人臉辨識方法。一般而言，PCA用於人臉辨識時，對於光照變化、透視效應、水霧干擾、人臉遮蔽等問題相當敏感，本研究提出一種強化影像法搭配PCA以解決這些問題。首先利用AdaBoost演算法將人臉由灰階影像中偵測框出，採用測邊法、水平/垂直投影法等取得眼線進行影像定位及影像正規化，在取得全臉與臉局部影像做為特徵影像後，再運用影像等化法搭配多層Retinex影像處理法(或稱加強型MSR，EMSR)，以強化影像特徵，提升利用PCA之人臉辨識率。PCA的人臉辨識法分成訓練與測試兩階段；在訓練階段，利用上述所提方法建立全臉及臉局部區域特徵之特徵空間與特徵資料庫；在測試階段，對輸入測試影像進行辨識時，將主元件分析法處理後之特徵影像投影在訓練階段所建立之特徵空間中，並與各個特徵資料庫比對，以最小歐氏距離作為初步比對結果，再將這些各個的初步結果依其相似程度指數配以權重，最後計算評定辨識結果。實驗結果證明本研究所提之方法，較一般PCA更能增強辨識率效果；最後本研究將所提方法實際植入ARM嵌入式系統中，提供家中成員辨識，未來可廣泛應用於智慧家庭中。

關鍵字

PCA ； 人臉辨識 ； AdaBoost ； Retinex ； 影像處理 ； EMSR

並列摘要

The PCA methodology is a useful method for real time face recognition of facilities with an embedded system at smart home in the future. However, PCA methodology is sensitive to illumination variation, perspective effect, vapor effect, and facial occlusions. In this study, we try to propose a systematic method to solve these well-known difficult problems. First, the face detection approach takes AdaBoost of Haar-like features to extract the face and then facial features as recognition features from gray image. After these images are aligned and normalized, if necessary, a proposed method called EMSR, combined histogram equalization and multi-scale retinex is applied to remove the influence of illumination variation. Then, PCA algorithm is used to compile eigenvectors and database of feature images of training samples. In recognition stage, while a testing face image is tracking, its corresponding feature images and eigenvectors can be found. Then, the Euclidean distances between the eigenvectors of the tracking image and all training samples can be used to determine the recognition result. Furthermore, we use the weighted results to determine who the test image is. The experimental results demonstrate that the proposed method can improve the recognition rate. Finally, we implant the proposed method in ARM embedded system for verification and prove it is suitable at smart home for familial member recognition.

並列關鍵字

PCA ； Face recognition ； AdaBoost ； Retinex ； Image Procession ； EMSR

參考文獻

[1] K. Sobottka and I. Pitas, "Extraction of facial regions and features using color and shape information", in Proc. 13th International Conference on Pattern Recognition, Vienna, Austria, Aug. 1996, pp. 421-425.

[2] Nagao, K., "Face recognition by distribution specific feature extraction", IEEE Conference on Computer Vision and Pattern Recognition, Proceedings, vol. 1, pp. 278-285, 2000

[3] G. Yang and T.S. Huang, “Human Face Detection in Complex Background”, Pattern Recognition, vol. 27, no. 1, pp. 53-63.

[4] Brunelli, R. and T. Poggio, Face recognition: features versus templates. 1993. 15(10): p. 1042.

[5] S. Baker, I. Matthews and J. Schneider, “Automatic construction of active appearance models as an image coding problem”, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 26, no. 10, pp. 1380-1384, 2004.