This paper presents a novel corner detection method for gray level images based on log-Gabor wavelet transform (WT). The input image is decomposed at multiscales and along multi-orientations. The magnitudes of the decomposition are formulated into the second moment matrix. The smaller eigenvalue of the second moment matrix is used as the "cornerness" measurement. Compared with the most famous Harris detector, SUSAN detector and the recently published detector - Gabor wavelet transform based detector, the proposed method shows good localization and single response to the higher order corner structures. The simulation results also shows the higher detection rate of the proposed method.

This paper presents two novel corner detection methods for gray level images based on log-Gabor wavelet transform. The input image is decomposed at multiscales and along multi-orientations. In the first algorithm, the magnitude along the direction that is orthogonal to the gradient orientation represents the "cornerness" measurement. Using this detector, corners are detected and localized accurately. Meanwhile, this method provides us the magnitudes and orientations along the principal axes corresponding to the directions of the local eigenvectors of the local changes. This information is useful when we deal with the affine transform applications in matching problems. The second proposed method is based on log-Gabor wavelets and second moment matrix. The input image is decomposed by the log-Gabor wavelets at multiscales along multi-orientations. Then the components at different scales and orientations are projected onto the x and y axis and formulated into the second moment matrix. Finally, the smaller eigenvalue of the second moment matrix is used to detect corner points. Compared with the most famous Harris detector and the recently published Kovesi's detector, the proposed methods provide better performance while retaining the good characteristics, such as robustness to noise, isotropic responses etc. Furthermore, the proposed methods address three problems existing in Harris detector. First, Harris method is based on single scale information. The proposed methods are based on the multiscales instead of single scale. Second, the derealization problem existing in Harris detector is minimized by utilizing good localization property of log-Gabor wavelets. Third, Harris detector has certain problems to detect corners of higher order structures. Due to the multi-orientation decomposition of the proposed methods, higher order structures are well computed. Consequently, the proposed methods detect corners of higher order structures very well. The proposed methods are first compared with Harris method and Kovesi's method in a subjective way. Then a classical stereo matching system is adopted to evaluate the corner detectors in an objective way. Both of the evaluations demonstrate better performance of the proposed methods.

A novel framework called 2D Fisher discriminant analysis (2D-FDA) is proposed to deal with the small sample size (SSS) problem in conventional one-dimensional linear discriminant analysis (1D-LDA). Different from the 1D-LDA based approaches, 2D-FDA is based on 2D image matrices rather than column vectors so the image matrix does not need to be transformed into a long vector before feature extraction. The advantage arising in this way is that the SSS problem does not exist any more because the between-class and within-class scatter matrices constructed in 2D-FDA are both of full-rank. This framework contains unilateral and bilateral 2D-FDA. It is applied to face recognition where only few training images exist for each subject. Both the unilateral and bilateral 2D-FDA achieve excellent performance on two public databases: ORL database and Yale face database B.

This paper addresses the small sample size (SSS) problem in linear discriminant analysis (LDA) utilizing a so called 2D Fisher discriminant analysis (2D-FDA) algorithm. As opposed to traditional LDA-based approaches, 2D-FDA is based on 2D image matrices rather than 1D vectors so the image matrix does not need to be transformed into a vector before feature extraction. The between-class scatter and the within-class scatter is constructed using the original image matrices. The advantage arising in this way is that the SSS problem existing in traditional linear discriminant analysis does not occur any more. To test the performance of 2D-FDA with small number of training samples, a series of experiments are conducted on two public databases: ORL and Yale face database B. In both two trials, the 2D-FDA outperforms the other linear subspace methods when there are only very limited training images for each subject.

A two-dimensional principal component analysis (2DPCA) by J. Yang et al. (2004) was proposed and the authors have demonstrated its superiority over the conventional principal component analysis (PCA) in face recognition. But the theoretical proof why 2DPCA is better than PCA has not been given until now. In this paper, the essence of 2DPCA is analyzed and a framework of generalized 2D principal component analysis (G2DPCA) is proposed to extend the original 2DPCA in two perspectives: a bilateral-projection-based 2DPCA (B2DPCA) and a kernel-based 2DPCA (K2DPCA) schemes are introduced. Experimental results in face recognition show its excellent performance.

This paper proposes an improved Plessey corner detection method for gray level images using multiscale analysis. Plessey corner detector is well known for its good performance. But, as we understand, Plessey corner detection method has three drawbacks. Firstly, it works only in the spatial domain, so it can only detect corners belonging to a specific scale. On the other hand, the proposed detector works in the scale-space domain, thus, it can detect the corner points belonging to different scales. Secondly, three parameters are needed to be set manually in the original Plessey method, while the proposed method requires to set only one parameter. Thirdly, the delocalization is a well-known inherent problem for the Plessey corner operator. And it can be increased with the scale at which it operates. The proposed algorithm partially solves this problem by detecting the corners from smaller scale to larger scales. Our proposed multiscale scheme can also be applied to other spatial corner detectors to improve their performances. Better simulation results are shown and compared with the original Plessey and SUSAN corner detectors.

This paper presents a method that matches interest point features detected on two images taken from different view points. A new multi-scale Plessey corner detector (MPCD) is used to detect the interest points. The geometric constraint between two images is exploited as in [R. Deriche, et al., 1994]: the fundamental matrix is derived using least median of squares (LMedS) from an initial set of matches, and then it is used to guide new matches. However, we propose a new energy function that can approximate affine transformation in a more effective way. Then, the initial set of matches is derived by minimizing the energy function. As a result, our method can perform well even when the pose variation is large between the two images. We compare our method using the proposed MPCD against two standard corner detectors on image matching. Also we evaluate our proposed matching criterion against Zhang's. Our method gave better results in both experiments on real face images.

The paper presents a real-time head/face tracking system that uses synchronized disparity and intensity frames from stereo cameras. It integrates stereo, morphological watershed, gradient and feature extraction techniques to track the head/face. The face is tracked once the facial features, e.g. eyebrows, eyes, mouth and nostrils, are available. The head is tracked if the facial features are not available, e.g., when the person is distant from the stereo camera system or when the head is viewed in profile. Our approach automatically initializes without user intervention and can be re-initialized automatically if tracking of the 3D face pose is lost. The results on our face database and real image video sequences illustrate the effectiveness of the method.

A new approach for estimating 3D head pose form a monocular image is proposed. Our approach employs general prior knowledge of face structure and the corresponding geometrical constraints provided by the location of vanishing point to determine pose of human faces. Eye-lines and mouth-line are assumed parallel in 3D space, and the vanishing point formed by the intersection of the eyeline and mouth-line in the image can be used to infer 3D orientation and location of human face. Perspective invariance of cross ratio and harmonic range is used to locate the vanishing point stably. The robustness analysis of the algorithm with synthesis data and real face images are enclosed.

We present a novel approach, called the "one-circle " algorithm, for measuring the eye gaze using a monocular image that zooms in on only one eye of a person. Observing that the iris contour is a circle, we estimate the normal direction of this iris circle, considered as the eye gaze, from its elliptical image. From basic projective geometry, an ellipse can be back-projected into space onto two circles of different orientations. However, by using an anthropometric property of the eyeball, the correct solution can be disambiguated. This allows us to obtain a higher resolution image of the iris with a zoom-in camera and thereby achieving higher accuracies in the estimation. The robustness of our gaze determination approach was verified statistically by the extensive experiments on synthetic and real image data. The two key contributions are that we show the possibility of finding the unique eye gaze direction from a single image of one eye and that one can obtain better accuracy as a consequence of this.