Abbas Kouzani

Fractal face representation and recognition

This paper presents a face representation and recognition scheme based on the theory of fractals. Each face image is represented by its fractal model which is a small collection of transformation parameters. The transformation is carried out once for known face images. For recognition, the input face image is transformed and its fractal model is then compared against the database of fractal models of known faces. Feedforward neural networks are utilised to implement the compression and recognition parts. Some experimental results are presented. The maximum compression ratio obtained for the successful recognition of known faces was observed to be 89:1 (for a compression threshold of 0.002).

Wavelet packet face representation and recognition

A human face representation and recognition system, based on the wavelet packet method and the best basis selection algorithm, is proposed. Through conducting a set of experiments on three groups of training sets, the optimal transform basis (called the face basis), the best filter, and the best decomposition level are identified for the face image class. A face image is represented in a compressed form by its wavelet packet coefficients. For recognition, the compressed input face image is then compared against a database of compressed images of the known faces. The recognition results are presented.

Illumination-effects compensation in facial images

Based on the concepts of linear object classes and the principal components analysis, an illumination-effects compensation method is presented to transform an arbitrary-lit face image whose illumination effects are pre-determined, into a front-lit face image.

Multiresolution eigenface-components

This paper presents a face recognition system that imitates the multiresolution processing technique employed by the human visual system. In the proposed system, a different degree of importance is assigned to each part of a face image, and each region of the face image is processed with a different resolution. This proposed system reduces the computational complexity of the eigenface method, and achieves higher compression ratios and higher recognition rates in comparison with the eigenface method. Experimental results are presented and discussed.

Commonsense knowledge-based face detection

A connectionist model is presented for commonsense knowledge representation and reasoning. The representation and reasoning ability of the model is described through examples. The commonsense knowledge base is employed to develop a human face detection system. The system consists of three stages: preprocessing, face-components extraction, and final decision-making. A neural network-based algorithm is utilised to extract face components. Five networks are trained to detect mouth, nose, eyes, and full face. The detected face components and their corresponding possibility degrees allow the knowledge base to locate faces in the image and generate a membership degree for the detected faces within the face class. The experimental results obtained using this method are presented.

Face image matching using fractal dimension

A new method is presented in this paper for calculating the correspondence between two face images on a pixel by pixel basis. The concept of fractal dimension is used to develop the proposed non-parametric area-based image matching method which achieves a higher proportion of matched pixels for face images than some well-known methods.

Illumination invariant face recognition

Few of the face recognition methods reported in the literature are capable of recognising faces under varying illumination conditions. The paper discusses a method which can achieve a higher recognition rate than those obtained for existing methods. The novelty of this new method is the use of an embossing technique to process a face image before presenting it to a standard face recognition system. Using a large database of face images, the performance of the proposed method is evaluated by comparing it against the performances of three existing methods. The experimental results demonstrate the successfulness of the proposed method.

Quadtree principal component analysis and its application to facial expression classification

Presents a method called quadtree principal components analysis for facial expression classification. The quadtree principal components analysis is an image transformation that takes its name from the quadtree partition scheme on which it is based. The quadtree principal components analysis method implements a global-local decomposition of the input face image. This solves the problems associated with the existing principal components analysis and local principal components analysis methods when applied to facial expression classification.

Constructing a fuzzy grammar for syntactic face detection

This paper presents a structural face detection system. The proposed system consists of three stages; preprocessing, face-components extraction, and final decision-making. In the first stage, image conversion, colour operation, image restoration, and image enhancement are carried out. Face components are extracted in the second stage. A face model is defined, and a fuzzy grammar composed of octal chain codes is used to represent each of the seven face components. The practical limitations of this representation are considered. Structural components are detected, and the possibility degree that the extracted component is a real face component is determined. In the last stage, a commonsense knowledge base is employed for final evaluation. The detected face components and their corresponding possibility degrees allow the human face knowledge base to locate faces in the image and generate a membership degree for that face within the face class. The experimental results obtained using this method are presented.

Commonsense knowledge representation and reasoning with fuzzy neural networks

This paper highlights the theory of common-sense knowledge in terms of representation and reasoning. A connectionist model is proposed for common-sense knowledge representation and reasoning. A generic fuzzy neuron is employed as a basic element for the connectionist model. The representation and reasoning ability of the model is described through examples.