Sparse representation-based face recognition has gained considerable attention recently due to its robustness against illumination and occlusion. Recognizing faces from videos has become a topic of importance to alleviate the limit of information content in still images. However, the sparse recognition framework is not applicable to video-based face recognition due to its sensitivity towards pose and alignment changes. In this paper, we propose a video-based face recognition method which improves upon the sparse representation framework. Our key contribution is an intelligent and adaptive sparse dictionary that updates the current probe image into the training matrix based on continuously monitoring the probe video through a novel confidence criterion and a Bayesian inference scheme. Due to this novel approach, our method is robust to pose and alignment and hence can be used to recognize faces from unconstrained videos successfully. Moreover, in a moving scene, camera angle, illumination and other imaging conditions may change quickly leading to performance loss in accuracy. In such situations, it is impractical to re-enroll the individual and re-train the classifiers on a continuous basis. Our novel approach addresses these practical issues. Experimental results on the well known YouTube Face database demonstrates the effectiveness of our method.

The Cave Automatic Virtual Environment (CAVE) is a fully immersive Virtual Reality (VR) system. CAVE systems have been widely used in many applications, such as architectural and industrial design, medical training and surgical planning, museums and education. However, one limitation for most of the current CAVE systems is that they are separated from the real world. The user in the CAVE is not able to sense the real world around him or her. In this paper, we propose a vConnect architecture, which aims to establish real-time bidirectional information exchange between the virtual world and the real world. Furthermore, we propose finger interactions which enable the user in the CAVE to manipulate the information in a natural and intuitive way. We implemented a vHealth prototype, a CAVE-based real-time health monitoring system, through which we demonstrated that the user in the CAVE can visualize and manipulate the real-time physiological data of the patient who is being monitored, and interact with the patient.

<p>Sparse representation-based face recognition has gained considerable attention recently due to its robustness against illumination and occlusion. Recognizing faces from videos has become a topic of importance to alleviate the limit of information content in still images. However, the sparse recognition framework is not applicable to video-based face recognition due to its sensitivity towards pose and alignment changes. In this paper, we propose a video-based face recognition method which improves upon the sparse representation framework. Our key contribution is an intelligent and adaptive sparse dictionary that updates the current probe image into the training matrix based on continuously monitoring the probe video through a novel confidence criterion and a Bayesian inference scheme. Due to this novel approach, our method is robust to pose and alignment and hence can be used to recognize faces from unconstrained videos successfully. Moreover, in a moving scene, camera angle, illumination and other imaging conditions may change quickly leading to performance loss in accuracy. In such situations, it is impractical to re-enroll the individual and re-train the classifiers on a continuous basis. Our novel approach addresses these practical issues. Experimental results on the well known YouTube Face database demonstrates the effectiveness of our method.</p>

No abstract

Many approaches have been taken towards the development of a compliant stereo correspondence algorithm that is capable of producing accurate disparity maps within a short period of time. There has been great progress over the past decade due to the vast increase in optimization techniques. Currently, the most successful algorithms contain explicit assumptions of the real world such as definitive differences in disparity among objects and constant textures within objects.This thesis starts by giving a brief description of disparity, along with descriptions of some common applications. Next, it explores various methods used in common stereo correspondence algorithms, as well as gives an in depth description and analysis of top performing algorithms. These algorithms are later used to compare with the proposed algorithm.In the proposed algorithm, frequency stereo correspondence in parallel with the traditional color intensity stereo correspondence is used to develop an initial disparity map. Frequency stereo correspondence is achieved using a winner-take-all block based Discrete Cosine Transiii form (DCT) to find the largest frequency components as well as their positions to use in disparity estimation. The proposed algorithm uses methods that are computationally inexpensive to reduce the computational time that plagues many of the common stereo correspondence algorithms. The proposed algorithm achieves an average correct disparity rate of 95.3%. This results in a disparity error rate of 4.07% compared to the top performing algorithms in the Middlebury website [1]; the DoubleBP, CoopRegion, AdaptingBP, and ADCensus algorithms that have error rates of 4.19%, 4.41%, 4.23%, and 3.97%, respectively. Additionally, experimental results demonstrate that the proposed algorithm is computationally efficient and significantly reduces the processing time that plagues many of the common stereo correspondence algorithms. iv First and foremost, I would like to express my sincere and utmost gratitude to my Masters supervisor, Dr. Ling Guan, for his guidance and support throughout my time at Ryerson University. His enthusiasm, inspiration, and advice he had provided me for my Masters was both delightful and productive. He is truly a role model: not only as an enthusiastic and productive researcher, but also someone with a kind heart who cares greatly for others.It was a blessing for me to be part of the Ryerson Multimedia Research Laboratory (RML).I gained both knowledge and friendship during my tenure at RML, working with my fellow researchers and students. I would like to specifically thank Naimul Khan, Xiaoming Nan, Fei Guo, Ziyang Zhang, Ning Zhang, and Kevin Tang for their friendship and advice they have given me during my time as a Masters student in RML. All of you have made my years in RML one of the most memorable time of my life.Last but not the least, I would like to thank my family and friends for their unconditional love, encouragement, and continual support for the past two years.

Many approaches have been taken towards the development of a compliant stereo correspondence algorithm that is capable of producing accurate disparity maps within a short period of time. There has been great progress over the past decade due to the vast increase in optimization techniques. Currently, the most successful algorithms contain explicit assumptions of the real world such as definitive differences in disparity among objects and constant textures within objects. This thesis starts by giving a brief description of disparity, along with descriptions of some common applications. Next, it explores various methods used in common stereo correspondence algorithms, as well as gives an in depth description and analysis of top performing algorithms. These algorithms are later used to compare with the proposed algorithm. In the proposed algorithm, frequency stereo correspondence in parallel with the traditional color intensity stereo correspondence is used to develop an initial disparity map. Frequency stereo correspondence is achieved using a winner-take-all block based Discrete Cosine Transform (DCT) to find the largest frequency components as well as their positions to use in disparity estimation. The proposed algorithm uses methods that are computationally inexpensive to reduce the computational time that plagues many of the common stereo correspondence algorithms. The proposed algorithm achieves an average correct disparity rate of 95.3%. This results in a disparity error rate of 4.07% compared to the top performing algorithms in the Middlebury website [1]; the DoubleBP, CoopRegion, AdaptingBP, and ADCensus algorithms that have error rates of 4.19%, 4.41%, 4.23%, and 3.97%, respectively. Additionally, experimental results demonstrate that the proposed algorithm is computationally efficient and significantly reduces the processing time that plagues many of the common stereo correspondence algorithms.

No abstract

Many approaches have been taken towards the development of a compliant stereo correspondence algorithm that is capable of producing accurate disparity maps within a short period of time. There has been great progress over the past decade due to the vast increase in optimization techniques. Currently, the most successful algorithms contain explicit assumptions of the real world such as definitive differences in disparity among objects and constant textures within objects. This thesis starts by giving a brief description of disparity, along with descriptions of some common applications. Next, it explores various methods used in common stereo correspondence algorithms, as well as gives an in depth description and analysis of top performing algorithms. These algorithms are later used to compare with the proposed algorithm. In the proposed algorithm, frequency stereo correspondence in parallel with the traditional color intensity stereo correspondence is used to develop an initial disparity map. Frequency stereo correspondence is achieved using a winner-take-all block based Discrete Cosine Transform (DCT) to find the largest frequency components as well as their positions to use in disparity estimation. The proposed algorithm uses methods that are computationally inexpensive to reduce the computational time that plagues many of the common stereo correspondence algorithms. The proposed algorithm achieves an average correct disparity rate of 95.3%. This results in a disparity error rate of 4.07% compared to the top performing algorithms in the Middlebury website [1]; the DoubleBP, CoopRegion, AdaptingBP, and ADCensus algorithms that have error rates of 4.19%, 4.41%, 4.23%, and 3.97%, respectively. Additionally, experimental results demonstrate that the proposed algorithm is computationally efficient and significantly reduces the processing time that plagues many of the common stereo correspondence algorithms.