有效的特徵擷取於人臉偵測與表情辨識之應用

Applications of Effective Feature Extraction in Face Detection and Facial Expression

盧迦弘(Jia-Hong Lu)

指導教授 ： 吳憲珠

國立臺中科技大學/資訊學群/資訊工程系碩士班/碩士(2011年)

https://doi.org/10.6826/NUTC.2011.00065

摘要

在本論文中，主要探討兩個在影像中的物件偵測及辨識之相關主題：人臉偵測與臉部表情辨識。一般來說，人臉偵測與臉部表情辨識為了達到有效的偵測以及辨識準確率，而使用了較為複雜的訓練技術。如支援向量機，類神經網路。但是卻忽略了整體的訓練過程式需要消耗大了的時間完成。因此，本論文針對人臉偵測的比對效率及臉部表情辨識的訓練過程做改進，並且依然能達到一定的偵測及辨識率。
本論文在第三章主要應用樣板比對的方式於人臉及眼部偵測。因為一個具代表性的樣板是可以應用於製造業中，並用來控制製造過程中各環節的品質控管，或是用於影像空間中的邊緣偵測，而這樣的匹配方式具有簡單且直覺性的特性存在。在偵測人臉之前，使用以HSL色彩空間為基礎的膚色切割的方式，將預定的人臉區域與背景分割出來，再從多張數位人臉影像中，取得平均的人臉樣板做為偵測過程的比對樣板。在偵測過程中，透過索柏(Sobel)邊緣偵測技術為基礎的影像前置處理，以解決光線影響人臉偵測的問題。從實驗效果顯示，本技術可有效地抵抗光線的變化並且能夠將人臉偵測出來。
第四章提出一個針對人臉表情差異性的方法，可偵測出人的異常表情。在進行表情偵測之前，在此同樣使用了索柏(Sobel)邊緣偵測技術於鄰近像素值及中心像素值，並產生特徵向量做為臉部特徵。在進行偵測過程中，本技術修改了豪斯多夫(Hausdorff)距離測量法對於影像相似度做計算，隨即運用一個自適性的門檻值，對人臉影像做異常表情的偵測。在實驗當中，本技術使用了公開的人臉資料庫作為測量依據，並且能有效地將事前定義的正常表情與異常表情區分出來。
在第五章中，本論文提出一個在彩色影像當中進行臉部表情辨識的技術。在彩色人臉特徵擷取方面，須先將影像的色彩空間從RGB轉為HSL，並將HSL色彩空間中的色調(Hue)做量化，再利用記錄鄰近像素的顏色資訊的方法做為特徵向量。最後使用第四章提出的豪斯多夫距離測量法並進一步做修改來進行辨識。在訓練方面，本技術只需一張個人的特定表情影像，即可針對此表情在彩色影像中進行偵測。在實驗結果當中，同樣的對於灰階影像進行測試，證明本技術可以有效的呈現人臉紋理，並且對於彩色影像也同樣能夠有不錯的辨識率。

關鍵字

膚色切割 ； 人臉及眼部偵測 ； 異常表情辨識 ； 豪斯多夫距離測量法

並列摘要

In this thesis, two major topics about image object detection and recognition, face detection and facial expression recognition are researched. Face detection and facial expression recognition usually require complex learning techniques, which are support vector machine (SVM) or artificial neural network (ANN) to achieve the accuracy of detection and recognition for efficiency. Therefore, the efficiency of face detection and the training procedure of facial expression recognition are improved and also have high accuracies of classification in this thesis.
The template matching method used for detecting face and eyes are described in Chapter 3. A representative template can be used in manufacturing, such as a part of quality control or edges detection on images. Before detecting faces and eyes, an HSL color space based skin color segmentation is used to divide the faces and background. Then, an average face template is obtained from several face images for template matching. In the detection phase, Sobel edge detection used to solve the problem of illumination variation. Experimental results show that the proposed scheme can effectively reduce the influence of illumination variation, and the faces and eyes of image are detected.
Chapter 4 presents a method based on dissimilarity of facial expression for detecting the unusual facial expression. Before detecting the facial expression, Sobel edge detection is also used to generate the feature vectors of face with neighboring and center pixels. In the detection phase, modifying Hausdorff distance is utilized to compute the similarity of images, and then an adaptive threshold is used for detecting the unusual facial expression. In experimental results, the proposed scheme could detect the predefined unusual facial expression in the public face database.
In Chapter 5, this thesis proposes a facial expression recognition scheme based on color image. For the color face feature extraction, the image is transformed from RGB to HSL color space, then taking the Hue of HSL for color quantification, and the color information of neighboring pixels is used to generate the feature vector. Finally, Hausdorff distance measurement is also applied to compute the similarity of images for facial expression recognition. In the training phase, it just needs one individual image of facial expression to detect the specific facial expression in color images. The experimental results reveals better performance in color images and the accuracy in gray-level images prove the face texture representation in the scheme is effectively.

並列關鍵字

Facial expression detection ； Facial expression recognition ； Hausdorff

參考文獻

[1] B. Ballarò, A. M. Florena, V. Franco, D. Tegolo, C. Tripodo, and C. Valenti, “An Automated Image Analysis Methodology for Classifying Megakaryocytes in Chronic Myeloproliferative Disorders,” Medical Image Analysis, Vol. 12, No. 6, 2008, pp.703-712.

[2] Biederman, “Recognition-by-components: A Theory of Human Image Understanding,” Psychological Review, Vol. 94, 1987, pp.115-147.

[3] C.Shan, S. Gong and P. W. McOwan, “Facial expression recognition based on Local Binary Patterns:A Comprehensive Study,” Image and Vision Computing, Vol.27 , 2009, pp.803-816.

[4] D.P. Huttenlocher, G.A. Klanderman and W.A. Rucklidge, “Comparing Images Using the Hausdorff distance,” IEEE Transaction, Pattern Analysis and Machine Intelligence, Vol.15, No.9, 1993, pp.850-863.

[6] G. R. S. Murthy and R.S.Jadon, “Effectiveness of Eigenspaces for Facial Expressions Recognition,” International Journal of Computer Theory and Engineering, Vol. 1, No. 5, 2009, pp.1793-8201.