Seok Cheol Kee

Face recognition using principal component analysis of Gabor filter responses

This paper addresses a new face recognition method based on principal component analysis (PCA) and Gabor filter responses. Our method consists of two parts. One is Gabor filtering on predefined fiducial points that could represent robust facial features from the original face image. The other is transforming the facial features into eigenspace by PCA, which is able to classify individual facial representations. Thus, the trained face model has some eigenvalues that can be derived from an ensemble matrix of given Gabor responses. In order to identify the faces, test images are also projected into eigenspace from the image space and compared to the trained face images in the same eigenspace. The basic idea of containing the PCA and Gabor filter is to overcome the shortcomings of PCA. When raw images were used as a matrix of PCA, the eigenspace cannot reflect the correlation of facial features well, because original face images have deformation due to in-plane, in-depth rotation and brightness and contrast variation. So, we have overcome these problems using Gabor filter responses as input. A Gabor filter has the robust characteristics of illumination and rotation. In addition, we confirmed the improvement of discrimination ability when Gabor responses had transferred to the space constructed by the principal components. The experimental results show that the proposed method achieves the remarkable improvement of recognition rate of 19% and 11% compared to conventional PCA method in SAIT dataset and Olivetti dataset respectively. Our method has the advantage in gallery DB size than the recognition method only using Gabor filter responses.

Component-based LDA Face Descriptor for Image Retrieval

We present a component-based face descriptor with LDA (Linear Discriminant Analysis) and a simple pose classification. Our algorithm has been developed to deal with face image retrieval in huge database such as those in internet environments. Such retrieval requires a compact face descriptor and an efficient recognition algorithm that is robust to variations in lighting and facial poses. Partitioning of a face image into components facilitates the development of an efficient and robust algorithm as follows. First, compensation for light and pose variations is much more easily done on individual components than on the whole image. Second, pose variation is compensated by classifying facial pose and aligning facial components. Finally, LDA is more effective at the component level which has simplified statistics than the whole image. Experimental results on MPEG-7 database show an impressive accuracy of our algorithm compared with conventional LDA methods.

Real-time normalization and feature extraction of 3D face data using curvature characteristics

The method of normalization and real-time feature extraction of 3D face data (range data) is presented. The step of normalization of range data is performed first using the symmetry of the defined facial section pattern and characteristics of changes of the pattern according to head rotations. Normalization of the data for head rotations can not only give strong constraints on the positions official features but also reduce the dimension of parameters used in the deformable template matching. Facial features are found in a range image, which is obtained by projection of the normalized range data, using the deformable templates of eyes, nose and mouth. For reliable feature detection, surface curvatures which can represent a local surface shape are used in this step. We define the energy functions of each template and the conditions of major control points using curvature information. Finally, the facial features are positioned in 3D space by back-mapping to the original range data. The back-mapping is the inverse process of getting the facial range image.

A fast eye localization method for face recognition

We introduce a fast, robust, accurate eye localization algorithm. Detecting and normalizing human faces from live video streams is the first crucial step in a face verification/recognition system. The accuracy and robustness affect the performance of the following face registration and classification. To localize face regions properly, we detect eye corners by using a corner detector and Gabor wavelets. First, by applying a corner detector in skin color regions, we dramatically reduce the candidate regions for eye corners. Extracted features are represented in a semilocal manner to increase discrimination. Then, in the set of the reduced candidates, a robust feature decision algorithm based on Gabor response analysis gives accurate eye corner locations. Experimental results on real images are presented.

Speaker detection and tracking at mobile robot platform

In this paper, we present a novel method to detect and track multiple people for an audio-visual speaker localization system using an uncalibrated camera and an 8 microphone array. We propose an object oriented speaker detection and tracking algorithm that uses a combination of skin color and human upper body part appearance information. Candidate regions of the input image are extracted from the color transform that is modeled by a 2D Gaussian function in a normalized RGB color space. We classify the humans and other objects which have a similar skin color region from the geometric structure and Hausdorff distance from the human shape. We decide the location of the speaker from audio and visual information and the robot tracks the speaker using CAMShift. The experimental results show that the proposed algorithm initializes and tracks the speaker automatically in various environments such as an image with skin color noise and complex background.

International standardization on face recognition technology

This paper details the international standard works on biometrics, especially face recognition, for example, which groups have an interest, what kind of works has been done, and why the standard works are needed Moreover, the history of MPEG-7 Advanced Face Recognition descriptor and current situations are described.

Level set methods, distance function and image segmentation

In the study of level set methods, several significant problems were neglected all along, such as the existence, uniqueness and singularities of level set methods. In this article we give the proof that in a neighborhood of the initial zero level set, for the level set equations with the restriction of distance function, there exists a unique solution, which must be the signed distance junction with respect to the evolving surface. We also present the analysis of singular points effect on level set evolution and give an adaptive narrow banding algorithm. The detailed numerical analysis and a simplified definition for singular points are presented. We give an adaptive narrow banding algorithm, which avoids the singular points and is proved to be robust and efficient in segmentation of CT data and synthesized images.

A novel heat kernel based Monte Carlo localization algorithm

A novel heat kernel based Monte Carlo localization (HK-MCL) algorithm is presented to solve the degeneracy problem of conventional Monte Carlo localization: real-time global localization requires the number of initial samples to be small, whereas global localization may fail if the number of initial samples is small. The degeneracy problem is solved by an optimization approach called heat kernel based perturbation (HK-perturbation), which moves the samples towards the high likelihood area. HK-perturbation integrates the average local density and importance weight of samples to determine each samples perturbation probability. The strategy improves simulated annealing algorithm via an obvious form of temperature, both in time and space, with respect to average local density and importance weight of samples. Systematic empirical results in global localization based on sonar illustrate superior performance, when compared to other state-of-the-art updating of Monte Carlo localization.

Discriminant Analysis by Locally Linear Transformations

We present a novel discriminant analysis learning method which is applicable to non-linear data structures. The method can deal with pattern classification problems which have a multi-modal distribution for each class and samples of other classes may be closer to a class than those of the class itself. Conventional linear discriminant analysis (LDA) and LDA mixture model can not solve this linearly non-separable problem. Several local linear transformations are considered to yield locally transformed classes that maximize the between-class covariance and minimize the within-class covariance. The method invloves a novel gradient based algorithm for finding the optimal set of local linear bases. It does not have a local-maxima problem and stably converges to the global maximum point. The method is computationally efficienct as compared to the previous non-linear discriminant analysis based on the kernel approach. The method does not suffer from an overfitting problem by virtue of the linear base structure of the solution. The classification results are given for both simulated data and real face data.

Facial feature tracking by robust face segmentation and scalable rotational BMA

We proposed an algorithm for the tracking of facial feature points based on the block matching algorithm (BMA) with a new shape of window considering the feature point characteristics and scale/angle changes of the face. The window used in the proposed algorithm is the set of pixels in the 8 radial lines of 0 degree(s),45 degree(s),... from the feature point, i.e. the window has the shape of cross plus 45 degree(s) rotated cross. This shape of window is shown to be more efficient than the conventional rectangular window in tracking the facial feature points, because the points and their neighbor are not usually the objects of rigid body. But since the feature points are usually on the edges of luminance or color changes, at least one of the radial line crosses the edge and it gives distinct measure for tracking the point. Also the radial line window requires less computational complexity than the rectangular window and more readily adjusted with respect to scale and angle changes. For the estimation of scale changes, the facial region is segmented at each frame using the normalized color, and the number of pixels in the facial region are compared.