Fast gradient-aware upsampling for cartoon video

In this paper, we propose a novel fast gradient-aware upsampling approach for cartoon animation. The advantage of our method is that we don´t need any prior knowledge and deconvolution stage to deblur the estimated result like existing methods. The low-resolution input video is initially upsampled to the desired high-resolution by a fast bicubic interpolation approach. After that, the gradient transformation is applied to the initially upsampled video frame. Finally, energy optimization is employed to recover additional fine details and structures. Our method is easy to implement and runs very fast. Moreover, in order to increase or decrease the sharpness of the reconstructed high resolution image, the user just needs to adjust one parameter γ. Thus, our approach is suitable for the task of cartoon video upsampling.