The paper presents a dictionary integration algorithm using 3D morphable facemodels (3DMM) for pose-invariant collaborative-representation-based faceclassification. To this end, we first fit a 3DMM to the 2D face images of adictionary to reconstruct the 3D shape and texture of each image. The 3D facesare used to render a number of virtual 2D face images with arbitrary posevariations to augment the training data, by merging the original and renderedvirtual samples to create an extended dictionary. Second, to reduce theinformation redundancy of the extended dictionary and improve the sparsity ofreconstruction coefficient vectors using collaborative-representation-basedclassification (CRC), we exploit an on-line elimination scheme to optimise theextended dictionary by identifying the most representative training samples fora given query. The final goal is to perform pose-invariant face classificationusing the proposed dictionary integration method and the on-line pruningstrategy under the CRC framework. Experimental results obtained for a set ofwell-known face datasets demonstrate the merits of the proposed method,especially its robustness to pose variations.
展开▼
