人臉影像自動校正及辨識系統

Abstract

隨著人臉影像技術發展，各項相關研究有如人臉偵測、對位、追蹤、辨識等領域均有進展。而隨著技術越趨穩定，許多需求也因應而生，如：應用人臉身分辨識技術之大樓智慧監控、企業員工績效查勤、門禁管制、海關通關人員檢視，犯罪人員檢搜、金融提款安全認證、電腦手機安全登入。而近年來人臉性別與年齡辨識也成為熱門研究主題，相關應用如百貨櫥窗智慧展示等需求也陸續開發中。 因此我們提出一套系統結合人臉自動偵測，影像對位，身分辨識、性別與辨識，可廣泛應用於眾多場合。本系統主要分成兩部分：第一階段：人臉影像正規化。使用者傳入影像、影片、或是藉由網路攝影機擷取畫面，系統便會自動進行人臉偵測，並且利用動態形狀模型（Active Shape Model），擷取人臉特徵點。利用這些臉上特徵點，將人臉影像精準對位，取出正規化人臉影像。第二階段：利用人臉特徵碼（Facial Trait Code），進行身分辨識；結合PCA（Principle Component Analysis），與SVM（Support Vector Machine）實行性別辨識。 人臉辨識演算法註冊單張影像，在無光線，表情變化之影響下，正確率可達95.3%，如果測試影像包含光線表情變化，正確率則為92.5%。性別辨識則使用3658 張男性與2720女性影像作為訓練資料，其正確率為94.8%. 系統單張從影像輸入，到辨識結果輸出，平均單人影像僅需0.65秒即可完成。

Face recognition can be applied to many different fields. For example, the building entrance security control, the criminal verification, and the identity of the network trade, etc. The gender and age recognition also can be used for department store. The digital screen can show different products with different genders people. We propose a fully automatic system that detects and normalizes faces in images and recognizes their identity and genders. There are two parts in this system. First step is face acquisition. We use Active Shape Model to find facial features, which are eyes, nose, and mouth. To boost the recognition accuracy, we correct the in-plane rotations of faces, and align faces based on estimated above positions of facial features. Second step is recognition. There are face recognition, gender recognition and age recognition. We use Facial Trait Code (FTC) to encode facial images for face recognition, and we combine the Principle Component Analysis (PCA) and Support Vector Machine (SVM) for gender recognition. Totally it cost only 0.65 seconds to process single face image from face detection to classification.