以即時影像處理人臉辨識系統之設計

Abstract

在本篇論文中，我們提出以即時擷取影像處理，利用人臉特有的特徵值，應用平均最小平方法及利用N個標準差來做提高辨識信賴度，來做為辨識率的演算法則。由於網路攝影機的性能，不管在高解析度的畫素及相機低失真度，越來越好，為了方便日後應用，本論文即應用具有USB的網路攝影機，來做影像處理，針對人臉獨有的生理特徵進行辨識並確認真實身份，首先分析正面臉，以眼睛及鼻孔中垂線，具有左半臉與右半臉內特徵對稱性及五官特徵，座落在何種區塊，並利用統計分析，應用於五官區塊搜尋，針對區塊內應用影像處理技術，擷取各區塊內的資訊，利用特徵擷取九個特徵值來做為人臉辨識的參考，因為九個特徵值的敏感度不同，在此思考利用控制系統識別，如FIR模型，求可得系統係數，將九個特徵值乘上系統的係數，可提高辨識程度，最後利用統計比較九個特徵值之間的變異，利用即時影像擷取到的平均值，應用平均最小平方法來做為與資料庫比對。利用N個標準差來區隔資料庫人員與非資料庫人員。

In this thesis, real-time image process using facial characteristic value is implemented the mean square and N-sigma methods to improve recognition reliability for human face recognition. Webcam’s performance will become more and more better in high resolution pixel and low distortion in the future. The USB of Webcam is used to capture the images and to recognize the human independence from physiological characteristics for individual identification. First, the front-face is analyzed by adopting the facial symmetry to research the locations and to profile the block areas of facial features. Appling the image processing technology and characteristic value of human face, the 9-term characteristic values are calculated and generated from the information of each block area. According to the identification theory of control system, the system coefficients of 9-term characteristic values might be weighted by using the signal process method, such as FIR model, to improve the recognition performance since each characteristic value possesses the different sensitivity. Finally, the means of real-time captured images are computed and compared the variances between the 9-term characteristic values. Furthermore, the recognition between the real-time image and human database is performed by applying the mean square method. The N-sigma method is used to distinguish between the member and the non-member in the human database.