| Summary: | 碩士 === 國立雲林科技大學 === 電機工程系碩士班 === 100 === The accuracy of face recognition system is inevitably error-prone to environmental interference, like illumination, light color, and shadow. So illumination/shadow/contrast normalization is the key point to the reliability of face recognition system. This thesis proposes a simple but practical method for illumination/shadow/contrast normalization, and improves face recognition system’s robustness to environmental illumination, irregular shadow, and light color variation.
In order to achieve all normalization of illumination, shadow, and contrast, this thesis reviews pros and cons of conventional methods, like histogram equalization, local binary pattern, and fast Homomorphic filter, and proposes a illumination/shadow/contrast normalization method, fast stretching Homomorphic filter. The proposed method is divided into two main steps. The first step is to normalize the illumination and shadow through fast Homomorphic filter, and the second step is to compress the contrast of the overall image through illumination stretching normalization and to enhance the contrast of regions of interest on human face simultaneously. Finally, the proposed fast stretching Homomorphic filter for illumination/shadow/contrast normalization is implemented and integrated on portable Android face recognition authentication system so as to accomplish mobile payment devices featuring 3-factor authentication and resisting various environmental interference.
|
|---|
