人臉辨識的研究迄今已有數十年的發展，其研究成果已能在機場的安檢系統、社區的門禁系統、ATM的認證或諸如許多商業化的NB影像認證等相關的安全辨識應用中發現其蹤影，但在上述的監控應用中，其辨識效果的好壞往往會受到表情、視角、光影、原始影像的解析度等因子所左右，也因為如此，一直以來都有許多相關的研究專門處理相對應的問題。 然而以現階段的研究來說，大多數的演算法對於具有角度變化的人臉辨識問題仍沒有一個很好的處理方法，因此本論文提出了一個兩階式的辨識方法，首先對輸入影像的視角做第一階段的預測，之後針對視角預測的結果進一步的辨識出其所屬的類別，這樣做的好處不但可以排除視角變化較劇烈的影像集，又可以減少影像比對的次數。 在實驗的部分，我們將漸進式的呈現出單一視角以及多重視角的實驗數據，實驗的最後我們可以得知，視角預測的前處理步驟可以對整個辨識系統的效能達到一定程度的提升。Face recognition technique has been developed for several years. The research results can be found in several applications, such as airport security system, access control system, ATM verification system, and surveillance system. However, the performance of identification result will be heavily affected by the factors of expression, pose, illumination and resolution of image. Hence, many relating algorithms were developed focusing on resolving these problems. Until now, most of the existing algorithms can still not fully resolve the pose problem in face recognition. In response to this need, we present a two stage identification method to improve the performance of face recognition system handling pose problem. In our proposed method, we first predict the pose variations for all input images. After that, we further classify the image class in the corresponding pose label set. The advantage of our proposed method can not only eliminate the difference in different pose sets but also reduce image matching time for recognition. Experimental results reveal that the proposed face recognition method with pose classifier can achieve better classification result.
