基於智慧型裝置之多使用者即時人臉辨識及權限控管研究

Abstract

人臉辨識是電腦視覺裡面一個重要的技術，近幾年由於身分認證，金融卡認證的需求日益增加，傳統的識別方法如密碼，身分證號碼存在可能的風險，而人臉辨識應用在智慧型手機上的需求更是日益漸增，像是身分認證，信用卡認證，手機解鎖，門禁管理，照片庫分類等等，而以手機上不同使用者登入來說，密碼以及圖形的輸入都存在著可以模仿的風險，所以生理特徵作為辨識的方法變得更為安全，也有其存在的必要性，現有的方法很多像是指紋，眼球虹膜，但在這些方法中，人臉辨識所需要的設備最為低價且最容易取得，也相對便宜。 本研究提出一個有效且快速的流程來辨識人臉，做為手機或平板電腦上的多使用者權限控管功能，亦可以應用到其他身分辨識的應用上，由於平板電腦的運算能力相對於一般電腦是較為薄弱的，所以本論文提出特徵擷取運算速度較快的noise-resilient LBP演算法，和特徵群聚法來解決平板電腦上記憶體不足的問題，研究方法共分成四個部分，一開始做人臉偵測找出人臉位置，再對該張人臉做影像前處理來克服不同光線的影響，提出noise-resilient LBP演算法進行特徵擷取，由於訓練集人臉特徵過多，因此本研究亦提出特徵群聚法來找出具有代表性的特徵，最後則是特徵距離相似度計算。

Face recognition is one of the important computer vision technology. Face recognition has many possible use, including classification of photos, unlocking/opening of doors, and factory access control. In this research, we proposed a real-time face recognition system for unlocking mobile phones/tablets and for granting access rights to the apps in the phone. Since smartphones and tablets have less computing power and computing resource as a computer, the published face recognition algorithm will not meet the real-time usage requirement. In this research, a fast noise-resilient LBP algorithm is proposed. The recognition procedure has four parts: first, localization of a human face; second, pre-processing of the localized face to reduce the uneven light source effect; third, the proposed noise-resilient LBP algorithm is used for feature extraction; and fourthly, feature clustering is performed to reduced the feature space. The experiments show that the proposed method is effective for real-time recognition of faces for up to 50 registered users in a mobile phone or tablet.