A Multiple-Faces Recognition System Applied to the Roll Call iOS App

[[abstract]]近年來配備多項感測元件與技術之智慧型個人行動裝置(如智慧型手機與平板電腦等)已逐漸普及。從「易用性」的角度研發對使用者友善的教學輔助相關應用，例如開發行動化、互動式教材等，達到簡單、易學、直覺操作與高效率的目標，已逐漸成為一種 App 開發趨勢。 本論文研究以開發一套利用行動裝置迅速完成課堂點名的教學輔助系統為目的，並能在平板電腦 Apple’s iPad上實現作為目標。首先發展基於人臉特徵配合影像處理程序以辨識出人臉之演算法，當中使用了Strong face check搭配Adaptive Boosting (Adaboost) 對 iPad 拍攝的相片 (可能包含多張學生人臉) 提取出所有人臉以達成人臉偵測 (Face Detection)，再利用 Speeded Up Robust Features (SURF) 演算法對偵測到的每一張人臉進行特徵提取與描述、與資料庫中修課學生樣本進行比對，以達成人臉辨識 (Face Recognition) 作為點名之結果。 以在平板裝置上能實現並有良好辨識率為前提，所提出的多人臉辨識演算法這項核心技術除了克服了複雜背景、攝像遠近等問題，研究當中更改善了人臉偵測技術進而降低誤判率，最後利用 iPad 上的各項裝置如攝像鏡頭、多點觸控螢幕與高性能處理器等進行實作。這些效能已由 iPad 實機應用獲得驗證。[[abstract]]In the recent years the personal intelligent mobile devices like smart phones and tablet computers which equip with many kinds of sensing elements and technologies have become more and more popular From the perspective of the usability developing user-friendly teach assist applications such as mobile and interactive learning materials to achieve easy learning intuitive operation and high-efficiency has become a trend of App development In this study we develop a teach assist system that can help teachers quickly call the roll in class using a mobile device such as Apple’s iPad This system involves two major image processing based techniques: face detection and face recognition First for the face detection we adopt the strong face check and the Adaptive Boosting(Adaboost) algorithm to detect faces within a picture (may contain multiple students’ faces) taken by the iPad Then for the face recognition we use the Speeded Up Robust Features(SURF) algorithm to extract/describe features for each detected face and match it with the samples in the database To be implemented on the iPad with great recognition rate the proposed face detection and recognition algorithm not only overcomes the complex background problem and the near-far issue but also improves the face detection technique to reduce false alarm rate The proposed system has been implemented on the iPad as an App which efficiently use the embedded camera multi-touch screen and the high-performance processor Some experiment results have also been presented to verify the efficacy of the proposed syste