Face Reconstruction and Recognition using Principal Components in RGB-D Space

• Main Authors: Po-Xun Wu, 吳柏勳
• Other Authors: Gee-Sern Hsu
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/41683615850847326704

碩士 === 國立臺灣科技大學 === 機械工程系 === 103 === Different from most RGB-D face recognition with RGB-D images available in both registration and recognition phases, this study considers RGB-D images only at the registration phase and the recognition is performed on RGB images. The quantization noise, often encountered when the subject is not close enough to the camera at registration, has not attracted much attention in the past, but is discussed and resolved by a proposed approach. The proposed approach exploits the principal components extracted from high resolution depth data as the basis for the reconstruction of the noise-corrupted facial depth. To deal with faces with eyeglasses, a reference facial model with eyeglasses is considered in the reconstruction. The performance of the proposed approach is evaluated on three public databases and the RGBDFaces that we made for studying the impacts of quantization noises.