Number of faces affect performance of rigid joint alignment

Good results have been obtained for face alignment in a single image. This is beneficial from use of auxiliary models or prior information. Compared with a single face image, there is much information available in multiple face images of a person. Rigid joint alignment will use the information as a whole. The paper mainly contain two works, i.e. constructing rigid joint alignment of multiple face images and testifying the effect of increasing number of images on performance of rigid joint alignment. When rigid joint alignment is carried out, an approach combining global rigid and local non-rigid is used for multiple images. To testify the effect of increasing the number of faces on performance of rigid joint alignment, tests are made in two public datasets, i.e. BioID and Pose face dataset. Results show that increasing the number of images does not improve performance of rigid joint alignment due to effect of some factors.