最近特徵線轉換 (Nearest Feature Line Embedding, NFLE)，是以 最近特徵線方法作為共變異數矩陣計算的空間轉換方式，在先前的研 究中，NFLE 已在實驗中顯示出它在人臉辨識上有不錯的效果，然而 在進行特徵線的分類計算時，大多採用歐氏距離，但歐氏距離對於每 個樣本在相同基準下進行計算，無法真實呈現出樣本的重要程度之 分。因此在本研究中，我們提出了 Fuzzy NFLE，利用加權式歐式距 離，將樣本依距離給予權重，距離近的樣本權重較大，距離遠的樣本 權重較小，如此可以模糊化樣本的類別資訊，使類別資訊能夠充份地 被利用，因此可以增加轉換空間的區別能力，我們以人臉辨識進行實 驗，實驗結果顯示 Fuzzy NFLE 辨識率可優於 NFLE。; Nearest Feature Line Embedding (NFLE) is a feature space transformation algorithm whose covariance matrix is obtained based on Nearest Feature Line. In previous studies, NFLE has successfully demonstrated its capability in face recognition However, the contribution of each training sample cannot be precisely extracted because NFLE is obtained based on Euclidean distance. To remedy this problem, fuzzy NFLE is introduced in this thesis. In our work, Fuzzy NFLE uses distance to evaluate each sample by assigning greater weight to closer sample in order to fully utilize the discriminative information of each sample with an eye to increasing feature space transfer capability. Experimental results demonstrate that the face recognition rate of Fuzzy NFLE performs better than NFLE.
