Image Resizing Based On High-order Generalized Barycentric Coordinates

Image resizing is widely used not only in the conversions between different screens with different resolution ratio but also images displayed by various software.At present,there are many algorithms distinguish the most important and the secondary areas of the image through saliency detection or user interaction.Those methods can avoid the distortion of the vital areas while satisfying the target aspect ratio through a specific scaling method.The image resizing method mainly discussed in this paper is a mesh-based image resizing algorithm.The existing deformation methods can only works on a single-topological triangular or quadrilateral mesh,using low-order polynomials.Although this kind of methods calculates fast,it often leads to unsatisfactory results due to its weak adaptive ability and low approximation order.In order to solve the problems,both a new method of representing the warping function based on high-order generalized barycentric coordinates and a contentaware automatic image resizing framework are proposed in the paper.The algorithm can express the deformation mapping as a piecewise smooth function of C0 continuity by the high-order generalized barycentric coordinates defined on the polygonal mesh,and then drive the image deformation by the polygonal mesh deformation.Firstly,we introduce the high-order generalized barycentric coordinates which can improve the representation ability of our warping function greatly and ensure the continuity between polygon meshes as well.Secondly,the saliency detection is introduced,and then the weight value of each pixel is determined by experiment.Then we introduce the detailed process of mesh division.Our input mesh is obtained through density map after short edge removal operation.Finally,by minimizing the energy function,the expression of the deformation function is determined.At last,the deformation framework based on the quadratic generalized barycentric coordinates is applied to a set of image resizing gallery,compared with 14 existing methods(3 of them are the latest methods and 11 of them are classic ones).We utilize the OQA scoring system to score the results of each method.The experimental results show that our content-aware image resizing algorithm achieves a better balance between resizing quality and mesh size,resulting in better results overall.In addition,the full-automatic framework is compatible enough to be applied to other generalized barycentric coordinates,saliency detection algorithm and deformation energy function.