Research Article
The Use of Selected Transforms to Improve the Accuracy of Face Recognition for Images with Uneven Illumination
@INPROCEEDINGS{10.1007/978-3-642-30419-4_21, author={Tomasz Orzechowski and Andrzej Dziech and Tomasz Lukanko and Tomasz Rusc}, title={The Use of Selected Transforms to Improve the Accuracy of Face Recognition for Images with Uneven Illumination}, proceedings={Mobile Multimedia Communications. 7th International ICST Conference, MOBIMEDIA 2011, Cagliari, Italy, September 5-7, 2011, Revised Selected Papers}, proceedings_a={MOBIMEDIA}, year={2012}, month={5}, keywords={Discrete Transforms Discrete Cosine Transform DCT Periodic Walsh piecewise Linear Transform PWL Periodic Haar piecewise Linear Transform PHL Image preprocessing Face recognition Illumination reduction Illumination normalization}, doi={10.1007/978-3-642-30419-4_21} }- Tomasz Orzechowski
Andrzej Dziech
Tomasz Lukanko
Tomasz Rusc
Year: 2012
The Use of Selected Transforms to Improve the Accuracy of Face Recognition for Images with Uneven Illumination
MOBIMEDIA
Springer
DOI: 10.1007/978-3-642-30419-4_21
Abstract
This paper presents new methods of the illumination normalization in images preprocessed for face recognition system. The main problem in statistical methods of face recognition is illumination. Different lighting conditions between photos taken indoor and outdoor may drastically decrease the level of correct classification. Variations of the illumination lie mostly in low-frequency band, so it is possible to use several transforms operating on frequency domain of an image. This approach is to truncate appropriate number of coefficients in frequency domain to minimize variations under different lighting conditions. This paper presents methods using transforms such as: Two Dimensional Discrete Cosine Transform type II (2D-DCT-II) and two Periodic Piecewise-Linear Transforms, such as: Periodic Haar piecewise Linear Transform (PHL) and Periodic Walsh piecewise-Linear Transform PWL. The main advantage of this approach is that, it does not require any modeling steps and it can be implemented in real-time face recognition systems.
