【正文】
ean weight vector of the pth class. The test image can be classified to be in class p when min( )piD ??, where p test pD W T??and i? is the threshold. 3. Modular PCA method The PCA based face recognition method is not very effective under the conditions of varying pose and illumination, since it considers the global information of each face image and represents them with a set of weights. Under these conditions the weight vectors will vary considerably from the weight vectors of the images with normal pose and illumination, hence it is difficult to identify them correctly. On the other hand if the face images were divided into smaller regions and the weight vectors are puted for each of these regions, then the weights will be more representative of the local information of the face. When there is a variation in the pose or illumination, only some of the face regions will vary and rest of the regions will remain the same as the face regions of a normal image. Hence weights of the face regions not affected by varying pose and illumination will closely match with the weights of the same individual39。 Turk and Pentland,1991。 Modular PCA。M significant eigenvectors are chosen as those with the largest corresponding eigenvalus. From these eigenvectors, the weights for each image in the training set are puted as p .( ) ,TiK K iW E I A i K? ? ? ( 3) Where KE’s are the eigenvectors corresponding to the 39。111 ,Mp jK p n jKKnT W p j?????? ?? ( 11) Next the minimum distance is puted as shown below: 39。M =20, . eigenvectors corresponding to the 20 maximum eigenvalues of the covariance matrix. The aim of our tests was to pare the two algorithms with varying pose and illumination face images and varying 39。因?yàn)槿四樀囊恍┠?部特性不會(huì)變化,甚至在姿態(tài),光照方向,面部表情變化時(shí),我們期望這種方法能夠適應(yīng)上述幾種的變量的變化。它是一種統(tǒng)計(jì)方法,人臉被表示為他們特征向量的一個(gè)子集,并因此被稱為特征臉 ( Sirovichand Kirby 1987; Turk and Pentland 1991;Moghaddam and Pentland 1997; Martinez, 2021; Graham and Allinson, 1998) 。我們提出了分塊 PCA 方法,這是傳統(tǒng) PCA 方法的一個(gè)延伸。第三節(jié)闡述分塊 PCA方法。 在人臉數(shù)據(jù)庫中的所有人臉被以很長 的向量進(jìn)行表示,而不是用普通矩陣進(jìn)行表示。這些圖像可以被表示為 2L 維的向量,或 維空間的一個(gè)點(diǎn)。從這些特征向量,每一幅圖像在訓(xùn)練組的分量通過如下計(jì)算出來 .( ) ,TiK K iW E I A i K? ? ? ( 3) 其中 39。當(dāng) min( )pD ??時(shí),這個(gè)測(cè)試圖像可以認(rèn)為是 p 類的,其中p test pD W T??和 i? 是閥值。 在這個(gè)方法中,每一幅在訓(xùn)練集中的圖像被分成 N 個(gè)小的圖像。通過特征向量計(jì)算權(quán)重如下所示: .( ) , , ,Tp n jK K p n jW E I A p n j K? ? ? ( 9) 其中 K取值為 1,2, … , 39。在訓(xùn)練集對(duì)應(yīng)的人臉類是最接近測(cè)試圖像。圖3a和 3b 顯示了一個(gè)人的一組分別用于訓(xùn)練和測(cè)試的圖像。取出一個(gè)人的 11 張圖像,只有 8 張用于訓(xùn)練,其他的三張用于測(cè)試識(shí)別率。圖 5 顯示了在不同數(shù)量特征向量下 PCA 和分塊 PCA 的識(shí)別率。據(jù)觀察分塊 PCA 算法提供了更好的識(shí)別率,它不需要檢測(cè)特定部位如眼鏡,鼻子和嘴。例如當(dāng) N=256 時(shí), 16 個(gè)特征向量是被考慮的。由于數(shù)據(jù)庫中所有圖像的大小都是 64*64 像素, N最大可以取到 4096,即一個(gè)子圖就是一個(gè)單個(gè)像素。對(duì)于 PCA 而言,識(shí)別率為 ,虛假識(shí)別率為 和 錯(cuò)誤拒絕率為 。 在光照變化 條件下,基于 PCA 的方法不是非常有效。 6 結(jié)論 分塊 PCA 方法 已經(jīng)提出了。 參考文獻(xiàn): [1] Chellappa ,R. Wilson, . Sirohey,. Human and machine recognition of faces: A survey. P。因此,在光照變化時(shí),人臉區(qū)域的權(quán)重不會(huì)受到變化的光照的影響,緊密與在正常情況下同一個(gè)的人臉區(qū)域權(quán)重相匹配。在 N=16 時(shí),使用 PCA 和分塊PCA 重構(gòu)一張測(cè)試圖像。像以前一樣,我們變化 N 從 4 到 4096,以觀察它對(duì)人臉識(shí)別的影響。訓(xùn)練和測(cè)試圖像按照 節(jié)所講進(jìn)行選擇。 當(dāng)子圖大小是小于或等于 4*4( N=256),可以從協(xié)方差矩陣得到的特征向量的數(shù)量將小于 20,因此協(xié)方差矩陣大小小于或等于 16*16。當(dāng) N=4,16,64,256,和 1024 已經(jīng)觀察到了類似的結(jié)果。 5測(cè)試結(jié)果 我們測(cè)試了不同數(shù)量特征向量的 PCA 和分塊 PCA 算法的性能。也有光源在中心,左邊,右邊的圖像。每個(gè)人的每個(gè)圖像都是在不同的姿態(tài)和正常的表情下采集的。39。我們把特征向量記為39。因此期望通過以下分塊 PCA 方法提高識(shí)別率。這個(gè)向量可以用來使測(cè)試圖像與預(yù)先確定的臉部類相匹配。平均臉定義為: M11A= IM ii?? ( 1) 每個(gè)臉偏離平均臉程度用向量 YiiIA?? 表示,協(xié)方差矩陣 C 為: M T11C= YM iii Y?? ( 2) 協(xié)方差矩陣的特征向量被計(jì)算出來,同時(shí)選擇 39。圖 1以圖形的方式闡釋了這個(gè)想法。把給定的圖像用被看成全局面部特征的特征圖像進(jìn)行擴(kuò)展。在這個(gè) PCA 方法中 PCA方法應(yīng)用在面部圖像的眼睛和鼻子。但是這種技術(shù)在光照和面部姿態(tài)變化非常大時(shí)并不十分準(zhǔn)確。 Chellappa 等人在 1995 年提交了一份關(guān)于統(tǒng)計(jì)方法,神經(jīng)網(wǎng)絡(luò)方法和特征方法的調(diào)查。這個(gè)算法與傳統(tǒng)的 PCA 算法相比提高了人臉圖像在光照方向和面部表情劇烈變化下的識(shí)別率。M , and there is not much improvement for 39。12, ,..., ME E E.The weights are puted from the eigenvectors as shown below: .( ) p, , ,Tp n jK K p n jW E I A n j K? ? ? ( 9) where K takes the values 1,2,…, 39。 Graham and Allinson,1998).PCA has also been used for handprint recognition (Murase et al.,1981), humanmade object recognition (Murase and Nayar, 1995), industrial robotics (Nayar et al.,1996), and mobile robotics (Weng,1996).But results show that the recognition rate is not satisfactory for pose variations exceeding 30176。附 錄 一、 英文原文 An improved face recognition technique based on modular PCA approach Rajkiran Gottumukkal Vijayan Abstract A face recognition algorithm based on modular PCA approach is presented in this paper. The proposed algorithm when pared with conventional PCA algorithm has an improved recognition rate for face images with large variations in lighting direction and facial expression. In the proposed technique, the face images are divided into smaller subimages and the PCA approach is applied to each of these subimages. Since some of the local facial features of an individual do not vary even when the pose, lighting direction and facial expression vary, we expect the proposed method to be able to cope with these variations. The accuracy of the conventional PCA method and modular PCA method are evaluated under the conditions of varying expression, illumination and pose using standard face databases. Keywords: PCA。 Martinez, 2021。M largest eigenvalues. We represent the eigenvectors as 39。 hence there are no rejections, only correct recognition or false recognition. It can also be observed from that the recognition rate is increasing in both PCA and modular PCA methods as we increase the value of 39。.Transactions on IECE J64D(3). [7] Murase, H., Nayar, S., 1995. Visual learning and recognition of 3D objects from . Computer Vision 14,5–24. [8] Nayar, ., Nene, ., Murase, H., 1996. Subspace methods for Robot (5),750–758. [9] Pentland, A., Moghaddam, B., Starner, T., 1994. Viewbased and modular eigenspaces