Back to results list
Show full item record
Please use this identifier to cite or link to this item:
|Title:||Face tracking and verification in video||Authors:||Choi, Wing-pong||Degree:||M.Phil.||Issue Date:||2008||Abstract:||The objective of this research is to develop novel algorithms for face tracking and verification in video. Our research focuses on the areas of facial feature representations and feature matching algorithms, both of which are efficient and effective for tracking and verification. The outputs from this research can be integrated to build a real-time face tracking and verification system. The automatic detection and tracking of human faces has many valuable applications, such as human computer interaction, visual surveillance, access control in special areas, etc. An accurate face tracker will definitely improve the performance of face recognition and other human activity analysis applications that are currently beyond the capabilities of current face tracking technology. We have proposed an effective face tracking algorithm based on the combination of shape and texture information. The edge map is used to represent the shape of a face, while the texture information is characterized by the local binary pattern (LBP). As the face patterns to be tracked in consecutive frames are highly correlated, accurate tracking can be achieved by searching for the shortest weighted feature distance between the face pattern and the possible face candidates. The weights of the shape and the texture can be adapted for real-time tracking. Both the edge map and the LBP can, to a certain extent, alleviate the illumination effect. Moreover, skin-color-like objects will not be falsely tracked as a face. Our proposed algorithm complements the AdaBoost face detection algorithm to form a multi-view face-tracking system. Experimental results show that our algorithm can track faces in varying poses (tilted or rotated) in real time. Beyond the detection or tracking of faces, recognition is performed to verify the identity of the tracked faces for visual surveillance. In order to make a practical surveillance system, the face recognition algorithm must be both accurate and efficient. We have proposed that simplified Gabor wavelets (SGWs) be applied to face recognition. Gabor wavelets (GWs) are commonly used to extract texture features for various applications, such as object detection, recognition and tracking. However, extracting Gabor feature is very computationally intensive, so Gabor-related methods are impractical for most real-time applications. This has inspired us to investigate a simplified version of Gabor wavelets and an integral image based algorithm for extracting Gabor-like feature efficiently. We have evaluated the performance of the SGW feature for face recognition. Experimental results show that using SGWs can achieve a performance level similar to that of using GWs, while the runtime for feature extraction by using SGWs is, at most, 4.39 times faster than that using GWs implemented by using the fast Fourier transform (FFT). Extracting the SGW features of 5 different scales and 4 different orientations from a 64x64 image takes only 16.09ms. This is a very encouraging result which allows the use of SGW features for real-time applications.||Subjects:||Hong Kong Polytechnic University -- Dissertations.
Human face recognition (Computer science)
|Pages:||xviii, 119 p. : ill. ; 30 cm.|
|Appears in Collections:||Thesis|
View full-text via https://theses.lib.polyu.edu.hk/handle/200/1944
Citations as of May 15, 2022
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.