基于三维多分辨率形变模型的人脸识别算法研究-计算机应用技术专业论文.docxVIP

AbstractAbstract Abstract Abstract The research work of this paper focuses on modeling and recognizing faces from different poses and illuminations In our researches for face modeling，we present a novel resampling based method for pixel-wise correspondence between 3d faces and a new category for image analysis．This paper presents a novel method for recognizing faces in images taken from difierent directions and under different illuminations．This method iS based on combination of a multi-resolution 3D morphable model and Fisher’S Linear Discrimination Analysis(FLDA)． A new technique is proposed for modeling 3D textured faces in different resolutions． Resampling based method，which advances a particular even—partition procedure including uniform segmentation and iso-curves initialization，is used to construct pixel—wise correspondence in this technique．It uniformly pararneterizes original grids of different prototypes through polygonal surfaces resampling，such that original geometry is maintained． The model Call also adapt various conditions such as head orientation，illumination and other parameters．However,instead of the time consuming computation of dense face model matching for each iteration step，a multi-resolution model is used to adapt images from tow to high resolutions．Both fitting coefficients and other parameters，acquired in the current resolution Can be set as the initial va]ue in the next resolution．This strategy would improve greatly the efficiency and the speed for the image analyzing procedure．FLDA is used to analyze the reconstructing results obtained from face images across different poses and illuminations．It produces we separated classes even under the condition ofsevere variation in poses and illuminations The even-gridded textured 3D models have been shown in different resolutions．With this multi-resolution model，we also demonstrate how to refine the method for 3D face reconstruction from a single image．Combined with FLDA，experimental results show