以次空間技術進行雜訊人臉影像辨識之研究和應用

A Study of Noisy Face Recognition using the Subspace Methods and its Applications

葉子銘(Tzu-Ming Yeh)

指導教授 ： 林文暉

崑山科技大學/資訊科技學院/資訊管理研究所/碩士(2011年)

摘要

近年來人臉辨識一直是極為熱門的研究領域，相當多具有創新性且可行的技術已被發展出來，這些技術主要基於不同的次空間特徵擷取方法，特徵擷取方法可概略分類成線性人臉特徵擷取技術和非線性人臉特徵擷取技術，線性人臉特徵擷取技術如主分量分析法(Principle Component Analysis，PCA)，二維主分量分析(Two-Dimensional PCA，2DPCA)，線性識別分析(Linear Discriminant Analysis，LDA)，獨立分量分析法(Independent Component Analysis，ICA)等，非線性人臉特徵擷取技術有核主分量分析(Kernel Principle Component Analysis，KPCA)，局部線性嵌入法(Locally Linear Embedding，LLE)，等構映圖(isometric feature mapping，ISOMAP)等。所有不同的臉部特徵擷取技術都具有其不同的潛在技術特性，但不外乎針對資料的維度降維(Dimensionality Reduction)、統計特性(Statistic Property)、非線性 (Nolinear)等特性進行特徵擷取以提升臉部辨識的效率和準確率。以上所提的技術及其衍生技術都有其擅長之處，然這些技術論文所探討的範圍很少涉及雜訊人臉影像，而在真實的人臉辨識應用時人臉影像很容易受到雜訊感染，因此本論文主要著重於既有的人臉次空間特徵擷取技術應用在雜訊影像的探討分析，探討臉部影像受到高斯雜訊、胡椒鹽雜訊感染時其效率和準確率的變化，同時根據多辨識器(Combining classifiers for Face Recognition)可提升準確率的優越特性下結合各辨識方法的特性進一步發展一個多辨識器的辨識法，以提升臉部辨識率。另外，論文中也提出一個臉部辨識的應用系統--家庭相簿。
本論文共分成五章，第一章介紹論文動機目的、真實世界臉部影像雜訊分析、人臉識別的基本方法與研究流程，第二章介紹各次空間人臉特徵擷取技術及提出多辨識器的辨識法，第三章實驗展示次空間特徵擷取法抗雜訊能力分析，第四章提出人臉辨識的應用—家庭相簿，此應用結合人臉辨識技術和族譜概念發展出一套可以由相片查詢親屬關係和相關資料的資訊系統，第五章討論和結論。

關鍵字

主分量分析法 ； 二維主分量分析 ； 獨立分量分析法 ； 線性識別分析

並列摘要

In the recent years, human face recognition has become a very popular field of research. There are a considerable number of innovative and practical techniques developed in human face recognition. These recognition techniques mainly based on different subspace feature extraction method that can be roughly classified into linear and non-linear. The linear feature extraction techniques such as the Principle Component Analysis, PCA, Two-Dimensional PCA, 2DPCA, Linear Discriminant Analysis, LDA and Independent Component Analysis, ICA etc. The Non-linear feature extraction techniques such as the Kernel Principle Component Analysis, KPCA , Locally Linear Embedding, LLE and isometric feature mapping, ISOMAP etc. All the facial feature extraction methods have their characteristics and potential for face recognition. The goals of these face feature extraction methods that can provide dimensionality reduction and statistical evaluation and nonlinear feature extraction of data can improve recognition efficiency and accuracy. Although, technologies mentioned above and its derivatives have advantages to face recognition, there have been rarely applied to the face image with noise. Unfortunately, in the real world application the face images were usually contaminated with noise when the faces were took. Therefore, in the thesis we will focus on the analysis of recognition efficiency and accuracy change when applying the well-known existing subspace feature extraction technology to the noisy images contaminated by different common noisy types included Gaussian noise and salt noise. We also provide a multi-classifier classifier based on the concept in Combining Classifiers for Face Recognition to achieve the best recognition accuracy. In addition, in the thesis also propose a face recognition application system - the family photo album. This application was developed by combining face recognition technology and genealogy concept that can provide the face image query and family relation query.
This thesis is divided into five chapters. The first chapter is the introduction included the motivation and the purpose of the study. The second chapter describes the various subspace technologies for facial feature extraction and the concept of the multi-classifier classifier. The third chapter shows the analysis of the subspace feature extraction for noisy face image, the fourth chapter describes the application of face recognition - the family photo album. The five chapter is our discussions and conclusions.

並列關鍵字

Principle Component Analysis ； Two-Dimensional PCA ； Linear Discriminant Analysis ； Independent Component Analysis

參考文獻

[30] 黃永昌、林文暉, 雜訊臉部影像辨識之研究, 崑山科技大學資訊管理系2009

腦部磁振造影像分析之研究”, 國立成功大學

[1] R. C. Gonzalez and R. E. Woods, Digital Image Processing. Pearson Education, Upper Saddle River, New Jersey.

[3] E. Abreu and S.K. Mitra, “A signal-department rank ordered mean (SD-ROM) filter- A new approach for removal of impulses from highly corrupted images”, 1995 International Conference on Acoustics, Speech, and Signal Processing, vol.4, pp2371 – 2374, May 9-12, 1995.

[4] H.L. Eng, and K.K. Ma, “Noise Adaptive Soft-Switching Mean Filter, IEEE Transactions on Image Processing”, vol.10, no.2, pp 242 - 251, Feb 2001.