Presenter_贾国琛.pptVIP

业精于勤,荒于嬉,行成于思,毁于随 * * * * * * * * * * 业精于勤,荒于嬉,行成于思,毁于随 Team 2:Firemonkey Presenter:贾国琛 业精于勤,荒于嬉,行成于思,毁于随 Outline 业精于勤,荒于嬉,行成于思,毁于随 * Introduction Background 2-D image-based methods are inherently limited. Using laser scanners. Motivation Given 3D scans of facial surfaces Develop metrics and mechanisms for comparing their shapes. [16] H. Li, S. M. Edwards, and J.-H. Lee. Measuring the intrusiveness of advertisements: scale development and validation 业精于勤,荒于嬉,行成于思,毁于随 * Introduction Related Work Detect a set of features and their locations , and use their relative placements to characterize the facial shape. Generate range images from 3D scans and utilize PCA-type techniques [16] H. Li, S. M. Edwards, and J.-H. Lee. Measuring the intrusiveness of advertisements: scale development and validation 业精于勤,荒于嬉,行成于思,毁于随 Facial Shapes S scanned face C level curve of F 业精于勤,荒于嬉,行成于思,毁于随 Comparing Shapes Use a shooting method to construct the geodesic path[7] 业精于勤,荒于嬉,行成于思,毁于随 Comparing Shapes The length of geodesic connecting curves Independent of rotation, translation, and scale in the x-y plane. 业精于勤,荒于嬉,行成于思,毁于随 Curve Extraction Data Collection FSU Database Notre Dame Database 业精于勤,荒于嬉,行成于思,毁于随 Curve Extraction mesh refinement and generation and masking of range images, level curve extraction from range images using imcontour in Matlab, computation of an angle function for each extracted curve, fitting a spline through its graph and resampling for arc-length parameterization, and, finally, computation of geodesic lengths between respective facial curves, and calculation of distances between facial surfaces. 业精于勤,荒于嬉,行成于思,毁于随 * Experiment Using five gallery faces per person, five facial curves per face, and geometric distance, we can achieve a recognition rate of almost 92 percent 业精于勤,荒于嬉,行成于思,毁于随 Thank s * * * by variability in imaging factors such as illumination and pose. for capturing surfaces of human faces are relatively invariant to illumination