Research On Facial Image Analysis Method Based On Geometry Features

Facial analysis plays very important role in many vision applications,such as authentication,surveillance and entertainments.There are two main problems in these tasks.One is to learn a corresponding relation between target and source.The very early works in 1990 s are mostly focus on estimating geometric deformations of facial landmarks to address this task.While in the past several years,more and more efforts have been made to directly learn an appearance regression for facial analysis.Though training regressions on controlled facial images can successfully capture the appearance variations,the performance of these appearance-based models are tightly related to the quantity and quality of the training data.Another one is to consider the facial analysis tasks as a image morphing task,i.e.the issue of deforming an image of a source object to that of a target.Previous works including barycentric coordinates and functional maps can hardly enforce shape consistency,especially for the objects with complex nested shape components.In this paper,for the two main problems,we propose two different methods to address the task.First,we develop a novel framework,named geometric correlated landmark regression(GCLR),to inherit the advantages but overcome limitations of appearance based and geometry based methods.Specifically,we first establish a landmark-to-landmark regression to estimate the geometry of facial images.By further incorporating a sparse coding term into the regression framework,we can successfully leverage the geometric correlations between test image and the shape dictionary,thus significantly enhance the geometry regression performance.Experimental results on various challenging facial datasets verify the effectiveness and efficiency of GCLR.Second,we leverage the conformal welding theory that maps 2D shapes(planar contours)to the automorphisms of the unit circle,named shape signatures.Conformal welding enables us to apply the Laplacian constraint to deformations in the signature space(or unit circle domain),which renders efficiency and flexibility.Additionally,we are able to fully reconstruct complex shape contours from deformed signatures,and hence generate the morphed images for target shapes.The experiments on complex shape contours and facial images,where multiple components exist,validate the effectiveness of the proposed approach.