Spectral convolutional sparse representation for mesh sequence

Learning the representation of mesh sequence is one of the most important topic in the computer graphics field. The representation can be used for deformation modeling, mesh editing and mesh preprocessing like filling holes, denoise. A meaningful representation can be learned from the sparse coding framework, which adopts numbers of basis to form a dictionary as the prior. Based on a review of the related approaches, we point out the challenges and drawbacks of current methods. Furthermore, we propose a new spectral convolution based sparse coding framework for mesh, which incorporate the mesh topology prior information with the sparse representation framework. Therefore, the proposed method can overcome the main drawbacks of canonical sparse coding framework. In the experiment, we compare the model with other methods. The result shows good performance in handling mesh data.