IMDSP-12.1

Face Recognition by Fractal Transformations
Teewoon Tan, Hong Yan (School of Electrical and Information Engineering, University of Sydney, Australia)

In this paper, we propose a new method for computerized human face recognition using fractal transformations. The popular use of fractal image coding has been for image compression. It is only recently that their uses for object recognition are being explored. We will show that by utilizing the intrinsic properties of block-wise self-similar transformations in fractal image coding we can use it to perform face recognition. The contractivity factor and the encoding scheme of the fractal encoder are shown to affect recognition rates. Using this method, an average error rate of 1.75% was obtained on the ORL face database.

IMDSP-12.2

Gesture Image Sequence Interpretation using Multi-PDM Method and Hidden Markov Models
Chung-Lin Huang (Electrical Engineering Dept., Univ. of Southern California, Los Angles, CA), Ming-Shan Wu (Electrical Engineering Dept., National Tsing-Hua University, Hsin-Chu, Taiwan)

This paper introduces a multi-PDM method and Hidden Markov Model for gesture image sequence interpretation. To track the hand shape, it uses the PDM model which is built by learning pattern of variability from a training set of correct annotated images. For gesture recognition, we need to deal with a large variety of hand-shape. Therefore, we divide all the training shape into a number of similar groups, with each group trained for an individual PDM shape model. Finally, we use the HMM to determine model transition among these PDM shape models. From the model transition sequence, it can identify the continuous gesture denoting one-digit or two-digit numbers.

IMDSP-12.3

Holistic Synthesis of Human Face Images
Joao L Maciel, Joao P Costeira (Instituto de Sistemas e Robotica - Instituto Superior Tecnico)

This paper presents a method to automatically synthesize human face images from holistic descriptions. We compactly represent the face set by a small set of prototypes, wich can be used in simple ways to generate controlled morphings. This becomes possible because separation of 2D-shape and texture provides a faithful, closed and convex representation of images, and smoothes the mappings between images and their properties. With this approach, the user watches an image being continuously morphed according to his indications, and the synthesized images always obey the natural physiognomic constraints.

IMDSP-12.4

Comparison of Approaches to Continuous Hand Gesture Recognition for a Visual Dialog System
Peter Morguet, Manfred Lang (Munich University of Technology, Germany)

Continuous hand gesture recognition requires the detection of gestures in a video stream and their classification. In this paper two continuous recognition solutions using Hidden-Markov-Models (HMMs) are compared. The first approach uses a motion detection algorithm to isolate gesture candidates followed by a HMM recognition step. The second approach is a single-stage, HMM-based spotting method improved by a new implicit duration modeling. Both strategies have been tested on continuous video data containing 41 different types of gestures embedded in random motion. The data has been derived from usability experiments with an application providing a realistic visual dialog scenario. The results show that the improved spotting method in contrast to the motion detection approach can successfully suppress random motion providing excellent recognition results.

IMDSP-12.5

AN EMBEDDED HMM-BASED APPROACH FOR FACE DETECTION AND RECOGNITION
Ara V Nefian, Monson H Hayes (Georgia Institute of Technology)

In this paper we describe an embedded Hidden Markov Model (HMM)-based approach for face detection and recognition that uses an efficient set of observation vectors obtained from the 2D-DCT coefficients. The embedded HMM can model better the two dimensional data than the one-dimensional HMM and is computationally less complex than the two-dimensional HMM. This model is appropriate for face images since it exploits an important facial characteristic: frontal faces preserve the same structure of ``super states'' from top to bottom, and also the same left-to right structure of ``states'' inside each of these ``super states''.

IMDSP-12.6

FACIAL FEATURES LOCALIZATION IN FRONT VIEW HEAD AND SHOULDERS IMAGES
Adnan M Alattar (Digimarc), Sarah A Rajala (College of Engineering, North Carolina State University)

The computerized process of locating human facial features such as the eyes, nose and mouth in a head and shoulders image is crucial to such applications as automatic face identification and model-based video coding. In this paper, a new model-based algorithm for locating these major features is developed. The algorithm estimates the parameters of the ellipse which best fits the head view in the image and uses these parameters to calculate the estimated locations of the facial features. It then refines the estimated coordinates of the eyes, mouth, and nose by exploiting the vertical and horizontal projections of the pixels in windows around the estimated locations of the features. The algorithm has been implemented and tested with over twelve hundred images, and simulation results indicate that the algorithm is robust to variations in subject head shape, eye shape, age, and motion such as tilting and nodding of the head.

IMDSP-12.7

Compensating for Variable Recording Conditions in Frontal Face Authentication Algorithms
Anastasios Tefas (Aristotle University of Thessaloniki), Yann Menguy (Matra Nortel Communications), Constantine Kotropoulos (Aristotle University of Thessaloniki), Gael Richard (Matra Nortel Communications), Ioannis Pitas (Aristotle University of Thessaloniki), Philip Lockwood (Matra Nortel Communications)

This paper addresses the problem of compensating for variable real recording conditions such as changes in illumination, scale differences, varying face position. It is well known that the performance of any face authentication/ recognition algorithm deteriorates significantly in the presence of the aforementioned conditions as well as the expression variations. The use of simple and powerful pre-processing techniques aiming at compensating for variable recording conditions prior to the application of any authentication algorithm is proposed. It is shown that such an approach overcomes indeed the image variations and guarantees an almost stable performance for the Morphological Dynamic Link Architecture developed within the European research project M2VTS.

< IMDSP-11

Last Update: February 4, 1999 Ingo Höntsch