| Summary: | 碩士 === 國立高雄應用科技大學 === 光電與通訊工程研究所 === 104 === Light variations for face recognition play a very important role. Poor quality image does not only exist excessive noise but also easy to lose information and reduce the overall system performance. This dissertation introduces a novel framework for solving the problem of lighting variation in color facial images. First of all, light detection using the singular value decomposition to sort out three different types of light variation of face images and using support vector machine for classification of even light, uniformly dark, and lateral lighting. Then, super resolution pixel clustering is adopted to classify face image into several local irregular blocks, and then the localized compensation is presented to overcome the variation of light. Finally, Taguchi method is performed with the analysis of control factors to optimize the design, and the best recognition rate. Experimental results with CMU-PIE, CMU-Multi-PIE, and CMU-Multi-PIE-poses show that our algorithm is robust for recognition under different illumination conditions.
|
|---|
