人臉辨識及表情辨識之整合設計

Abstract

本論文發展一套應用於機器人互動之人臉辨識與表情辨識之整合設計。人臉影像先經由主動外觀模型(Active Appearance Model, AAM)計算出人臉形狀以及紋理模型；接著對輸入的人臉影像進行改良式Lucas-Kanade 影像校正以找出人臉特徵，再利用AAM的人臉紋理模型建立出人臉紋理的特徵參數，並利用此特徵參數輸入倒傳遞類神經網路做辨識。在辨識過程中，我們提出一種整合設計，先經由人臉辨識找出使用者的身分，接著我們對於辨識出的已知使用者的表情資料庫做個人化表情辨識；由實驗結果可以看出，在使用BU-3DFE人臉表情資料庫做辨識，其人臉辨識成功率可達98.3%。接著使用個人化表情辨識的成功率為83.8%，相對於使用全體的表情辨識器之辨識率僅為69.6%，其成功率可大幅提升。

In this thesis, an integrated design of face and facial expression recognition system has been developed for robotic applications. First, facial image from camera is exacted to compute facial shape and texture model using active appearance model (AAM). Second, we use modified Lucas-Kanade image alignment algorithm to find facial features. Third, the texture model of AAM is used to construct facial texture parameters. These parameters are used to train a back propagation neural network (BPNN) for face and facial expression recognition. In recognition process, we first use face recognition to find user’s identity; then we use recognized user’s facial expression database to recognize his/her facial expression. In experiments based on BU-3DFE database, a face recognition rate of 98.3% has been achieved. The facial expression recognition rate of the proposed integrated method (using a personal facial expression classifier) is 83.8%. It is a great improvement compared with using conventional facial expression classifier of 69.6%.