Image Enlargement Based on Multi-scale Model Coupling of the Improved Forward-and-Backward Diffusion

In this paper,a novel image enlargement method based on the multi-scale space model is proposed to obtain high resolution images with good visual effects.To make an enlarged image meet the constraints of image degradation model,Laplace pyramids transform is combined with the degradation model.The original interpolated image which is zero-padding interpolation of the input image using Laplace pyramids transform is the component of low solution space.The component of high resolution space is derived from improved forward-and-backward anisotropic PDE diffusion through a spatial projection based on Laplace pyramids transform.The enlargement image is the sum of the two components.Compared with the original forward-and-backward anisotropic PDE diffusion,the proposed diffusion method enhances the edge,smoothes the edges at the same time.The experimental results demonstrate that the performance of our proposed method is prominent.Comparing with other existing PDE-based zoom methods,our method improves the accuracy of the edges and reduces the undesirable artifacts,such as ringing,block effects and edge distortion.Furthermore the method is faster with lower computational complexity.