It is evident that the process of face recognition, by definition, should be based on the content of a face. The problem is: what is a "face"? Recently, a state-of-the-art statistics-based face recognition system, the PCA plus LDA approach, has been proposed [1]. However, the authors used "face" images that included hair, shoulders, face and background. Our intuition tells us that only a recognition process based on a "pure" face portion can be called face recognition. The mixture of irrelevant data may result in an incorrect set of decision boundaries. In this paper, we propose a statistics-based technique to quantitatively prove our assertion. For the purpose of evaluating how the different portions of a face image will influence the recognition results, a hypothesis testing model is proposed. We then implement the above mentioned face ...

Abstract. Recognition in uncontrolled situations is one of the most important bottlenecks for practical face recognition systems. We address this by combining the strengths of robust illumination normalization, local texture based face representations and distance transform based matching metrics. Specifically, we make three main contributions: (i) we present a simple and efficient preprocessing chain that eliminates most of the effects of changing illumination while still preserving the essential appearance details that are needed for recognition; (ii) we introduce Local Ternary Patterns (LTP), a generalization of the Local Binary Pattern (LBP) local texture descriptor that is more discriminant and less sensitive to noise in uniform regions; and (iii) we show that replacing local histogramming with a local distance transform based similarity metric further improves the performance of LBP/LTP based face recognition. The resulting method gives state-of-the-art performance on three popular datasets chosen to test recognition under difficult

This paper proposes a gesture recognition method which uses higher order local autocorrelation (HLAC) features extracted from PARCOR images. To extract dominant information from a sequence of images, we apply linear prediction coding technique to the sequence of pixel values and PARCOR images are constructed from the PARCOR coefficients of the sequences of the pixel values. From the PARCOR images, HLAC features are extracted and the sequences of the features are used as the input vectors of the Hidden Marcov Model (HMM) based recognizer. Since HLAC features are inherently shift-invariant and computationally inexpensive, the proposed method becomes robust to changes of shift of the person's position and makes real-time gesture recognition possible. Experimental results of gesture recognition are shown to evaluate the performance of the proposed method. 1. Introduction Recently interest in gesture recognition has been rapidly increasing because of its broad range of applicability in nat...

I thank the committee chair and my advisor, Dr. Allen Hanson, for his support. I thank Dr. Bruce Croft for providing the environment to conduct this research. I thank Dr. Rod Grupen for his guidance over the years, and I thank Dr. Riseman for introducing me to Computer Vision. Prof. Haluk Derin’s sad loss precludes him from reading this thesis and I hope that it meets his standard. I thank very much Prof. Patrick Kelly for agreeing to be a reader and enjoy his insightful comments very much. I thank Dr. R. Manmatha for a fruitful collaboration, which lead to many of the foundation papers of this thesis. This dissertation would not be possible without the administrative, support and systems staff at the department, thanks especially to Sharon Mallory, Laurie Downey, Priscilla Coe, Kate Moruzzi and Pauline Hollister. I thank many colleagues from CIIR, Robotics, and Vision laboratories, for being a good source of intellectual development. I also thank the new students who have greatly enhanced my ability as a researcher. I thank Adam, Matt, Jahnava, Kate and Ben Jenkins for wonderful friendship and David Hirvonen for his. With immense gratitude, my thanks to Daniella for shining light and laughter on my life. Many thanks to my surrogate parents of New England, Marcelle and Paul Lipke, for their support and just the same to Pauline, Joe, Tory and Zeb Volpe, my family of Amherst, for putting up with me all these years!

image analysis. There are two major approaches: localfeature -based and image-vector-based. We propose a hybrid of these two approaches. Our method uses Higherorder Local Auto-Correlation (HLAC) features and Fisher weight maps. HLAC features are computed at each pixel in an image. These features are integrated with a weight map to obtain a feature vector. The optimal weight map, called a Fisher weight map, is found by maximizing the Fisher criterion of feature vectors. Fisher discriminant analysis is used to recognize an image from the feature vector. Our experiments on facial expression recognition demonstrate the effectiveness of Fisher weight maps for objectively quantifying the importance of each facial area for classification of expressions.

In this paper, a coarse-to-fine, LDA-based face recognition system is proposed. Through careful implementation, we found that the databases adopted by two state-of-the-art face recognition systems [4, 5] were incorrect because they mistakenly use some none-face portions for face recognition. Hence, a face-only database is used in the proposed system. Since the facial organs on a human face only differ slightly from person to person, the decision-boundary determination process is tougher in this system than it is in conventional approaches. Therefore, in order to avoid the above mentioned ambiguity problem, we propose to retrieve a closest subset of database samples instead of retrieving a single sample. The proposed face recognition system has several advantages. First, the system is able to deal with a very large database and can thus provide a basis for efficient search. Second, due to its design nature, the system can handle the defocus and 1 This work was partially sup...

This study investigates effective image features for characterization of local regions. We propose an extension of higher order local autocorrelation (HLAC) features. The original HLAC features are restricted up to the second order. They are represented by 25 mask patterns. We increase their orders up to eight and extract the extended HLAC features using 223 mask patterns. Large mask patterns are also created to support large displacement regions. They are used to construct multi-resolution HLAC features. The proposed method outperforms Gaussian Markov random fields, Gabor features, and local binary pattern operator in texture classification. 1

It is widely acknowledged that face recognition could play an important role in advanced video-based surveillance systems, mainly because it is non-intrusive and does not require people cooperation. Unfortunately, face recognition algorithms showed to suffer a lot from the high variability of environmental conditions (e.g., variations of lighting, face pose and scale). This currently limits their application to real video-surveillance tasks. Recently, fusion of multiple face recognisers has been proposed to improve the robustness of face recognition systems to environmental conditions variability. In this chapter, fusion of two well-known face recognition algorithms, namely, PCA and LDA, is proposed. Experimental results that confirm the benefits of fusing PCA and LDA allow drawing some preliminary conclusions about the role of the fusion of face recognition algorithms in video-based surveillance applications.

Automatic lipreading is important in various humancomputer interaction applications. Lipreading requires recognition not only of the mouth shape change but also of the appearance of the inner mouth (the teeth and the tongue). We have developed a lipreading system that can represent the changes of the mouth shape and the inner mouth appearance occurring throughout the input image sequence by producing a single 70 dimensional feature vector. This is achieved by generating the Cepstral coefficients of the pixel intensity change over time and arranging them as pixel intensities of a Cepstral image. Then the Higher Order Local Autocorrelation (HLAC) features are extracted from the Cepstral images and are used for classification. This paper explains the techniques used in the system, and reports on our feasibility study on the use of these techniques in automated lipreading.

This paper proposes a new scheme for multipurpose image recognition based on Higher-order Local Auto-Correlation (HLAC) features and factor analysis. First, HLAC features, which are inherently invariant under translation, computationally inexpensive, and additive, are extracted from the input images. Second, factor analysis is applied to the feature vectors so as to decompose the feature vectors as combinations of factors leant through supervised training examples. After the factorization, the input image is recognized by using the factor scores obtained through the least squares method. Experimental results show that the proposed method effectively enables the system to recognize images by acquiring effective factors that represent each object in the images without any need for segmentation or locating objects. 1.