Event Details
Presenter: Ana-Maria Sevcenco
Supervisor: Dr. Wu-Sheng Lu
Date: Fri, August 6, 2010
Time: 14:00:00 - 15:00:00
Place: EOW 230
ABSTRACT
ABSTRACT
Various face recognition techniques have been proposed in the past decades with a great deal of success. On the other hand, these methods often encounter difficulties when dealing with images captured under drastically different illumination conditions. Numerous tests have demonstrated that lighting variation is one of the bottlenecks in face recognition. Thus, there has been much work dealing with illumination variation for improved recognition rate. Generally, the algorithms in the literature can be classified into two categories: model-based and pre-processing-based. While the model-based approaches are theoretically ideal, when applied to real situations they require additional constraints and assumptions, and their computational cost increases accordingly, making them not very practical for real time systems. On the other hand, pre-processing-based techniques transform images directly without any assumptions or prior knowledge, being more commonly used in practical systems due to their simplicity and efficiency.
We provide an overview of several pre-processing techniques and propose a new pre-processing-based strategy based on perfect histogram matching (PHM). The PHM method is conceptually simple, easy to apply and computationally efficient, and it can be used as a pre-processing module in combination with face recognition processing techniques, for instance with conventional PCA. The experimental results show its efficiency in improving face recognition rates in comparison with several existing pre-processing techniques (e.g. WPCA, DCT-PCA and HE-PCA), and its versatility in combination with other pre-processing steps (e.g. WII).