Sang-Youn Lee

Face recognition using the embedded HMM with second-order block-specific observations

The paper is concerned with face recognition using the embedded hidden Markov model (EHMM) with second-order block-specific observations. The proposed method partitions a face image into a 2-D lattice type, composed of many blocks. Each block is represented by the second-order block-specific observation that consists of a combination of first- and second-order feature vectors. The first-order (or second-order) feature vector is obtained by projecting the original (or residual) block image onto the first (or second) basis vector that is obtained block-specifically by applying the PCA to a set of original (or residual) block images. A sequence of feature vectors obtained from the top-to-bottom and the left-to-right scanned blocks are used as an observation sequence to train EHMM. The EHMM models the face image in a hierarchical manner as follows. Several super states are used to model the vertical facial features such as the forehead, eyes, nose, mouth, and chin, and several states in the super state are used to model the localized features in a vertical face feature. Recognition is performed by identifying the person of the model that provides the highest value of observation probability. Experimental results show that the proposed recognition method outperforms many existing methods, such as the second-order eigenface method, the EHMM with DCT observations, and the second-order eigenface method using a confidence factor in terms of average of the normalized modified retrieval rank and false identification rate.

Face recognition using the second-order mixture-of-eigenfaces method

The well-known eigenface method uses an eigenface set obtained from principal component analysis. However, the single eigenface set is not enough to represent the complicated face images with large variations of poses and/or illuminations. To overcome this weakness, we propose a second-order mixture-of-eigenfaces method that combines the second-order eigenface method (ISO MPG m5750, Noordwijkerhout, March 2000) and the mixture-of-eigenfaces method (a.k.a. Gaussian mixture model (Proceedings IJCNN2001, 2001). In this method, we use a couple of mixtures of multiple eigenface sets: one is a mixture of multiple approximate eigenface sets for face images and another is a mixture of multiple residual eigenface sets for residual face images. Each mixture of multiple eigenface sets has been obtained from expectation maximization learning consecutively. Based on two mixture of multiple eigenface sets, each face image is represented by a couple of feature vectors obtained by projecting the face image onto a selected approximate eigenface set and then by projecting the residual face image onto a selected residual eigenface set. Recognition is performed by the distance in the feature space between the input image and the template image stored in the face database. Simulation results show that the proposed second-order mixture-of-eigenfaces method is best for face images with illumination variations and the mixture-of-eigenfaces method is best for the face images with pose variations in terms of average of the normalized modified retrieval rank and false identification rate.

An accurate COG Defuzzifier design using Lamarckian co-adaptation of learning and evolution

This paper proposes a design technique of optimal center of gravity (COG) defuzzifier using the Lamarckian coadaptation of learning and evolution. The proposed COG defuzzifier is specified by various design parameters such as the centers, widths, and modifiers of MFs. The design parameters are adjusted with the Lamarckian co-adaptation of learning and evolution, where the learning performs a local search of design parameters in an individual COG defuzzifier, but the evolution performs a global search of design parameters among a population of various COG defuzzifiers. This co-adaptation scheme allows to evolve much faster than the non-learning case and gives a higher possibility of finding an optimal solution due to its wider searching capability. An application to the truck backer-upper control problem of the proposed co-adaptive design method of COG defuzzifier is presented. The approximation ability and control performance are compared with those of the conventionally simplified COG defuzzifier in terms of the fuzzy logic controllers approximation error and the average tracing distance, respectively.

Hierarchical Shot Clustering for Video Summarization

Digital video is rapidly becoming a communication medium for education, entertainment, and a variety of multimedia applications. With the size of the video collections growing to thousnads of hours, efficient searching, browsing, and managing video information have become of increasing importance. In this paper, we propose a novel hierarchical shot clustering method for video summarization which can efficiently generate a set of representative shots and provide a quick and efficient access to a large volume of video content. The proposed method is based on the compatibility measure that can represent correlations among shots in a video sequence. Experimental results on real life video sequences show that the resulting summary can retain the essential content of the original video.

Scalable keyframe extraction using one-class support vector machine

In this paper, we present a scalable keyframe extraction method using one-class support vector machine. Keyframe extraction seeks to generate good images that best represent underlying video content and provide content-based access points. Criteria for good images play a major role for keyframe extraction process. Extracting good images can be viewed as detecting novel images among all the frames within a video. Therefore, keyframe extraction reduces to novelty detection problem. We describe how to efficiently solve the novelty detection problem using one-class support vector machine. We also present an algorithm of extracting keyframes in a scalable way so that one can access a video from coarse to fine resolution. We demonstrate the performance of our algorithm on several different types of videos.

Posed face image synthesis using nonlinear manifold learning

This paper proposes to synthesize posed facial images from two parameters for the pose. This parameterization makes the representation, storage, and transmission of face images effective. Because variations of face images show a complicated nonlinear manifold in high-dimensional data space, we use an LLE (Locally Linear Embedding) technique for a good representation of face images. And we apply a snake model to estimate face feature values in the reduced feature space that corresponds to a specific pose parameter. Finally, a synthetic face image is obtained from an interpolation of several neighboring face images. Experimental results show that the proposed method creates an accurate and consistent synthetic face images with respect to changes of pose.

Robust face-tracking using skin color and facial shape

This paper proposes a robust face tracking algorithm based on the CONDENSATION algorithm that uses skin color and facial shape as observation measures. Two independent trackers are used for robust tracking: one is tracking for the skin colored region and another is tracking for the facial shape region. The two trackers are coupled by using an importance sampling technique, where the skin color density obtained from the skin color tracker is used as the importance function to generate samples for the shape tracker. The samples of the skin color tracker within the chosen shape region are updated with higher weights. The proposed face tracker shows a robust tracking performance over the skin color based tracker or the facial shape based tracker given the presence of clutter background and/or illumination changes.

High quality VoD service architecture on broadband internet

In this paper, for broadband Internet users, high quality VOD (Video on Demand) service architecture including network and platform is presented. As the existing IP-network cannot provide a guaranteed bandwidth, a local server based network architecture is suggested and implemented. Local nodes are established to consider network traffic related to the number of users and geographical locality. Users are routed to the local node that can provide a guaranteed bandwidth. Also VoD service platform is designed to control traffic load balancing, which can distribute traffic load to local servers. The designed load balancing scheme is works well in actual environment. On the service architecture, commercial VoD (Video-on-Demand) service is now open to KT broadband Internet users (http://homemedia.megapass.net).