Study And Implementation To The Moving Least Squares Method For Image Deformation

Image deformation is an image process technique, which create smooth and realistic deformation effect through deformation function which defines by constraint relations. This technique is widely used in movie and cartoon, virtual reality, medical image processing field.In this paper, image deformations using different control features are discussed and compared. Then a point-based image deformation method using moving least squares (MLS) is studied. Three type transformations using Moving Least Square are compared:affine, similarity and rigid transformations.Based on the image deformation using MLS, We propose an improving method to MLS by adopting different weighting that can improve the deformation results, which lets customize deformation become possible. We also propose an efficient deformed image reconstruction method. This two improve points above can optimize the image deformation effect and increase the deformation effect. To test and verify the deformation effect of method, an image deformation application has been proposed based on the above-mentioned method on iOS platform, in which user can set up control points and move them to generate expected deformation. To reduce the time that our algorithm costs and improve the user experience, we propose an method using OpenGL ES technique to optimize the algorithm by concurrent computing in GPU using shader programing. Experiments show that image deformation speeds up at least one order of magnitude by the measure in this paper.