Complexity and quality evaluation of structure extrapolation methods within a fully automatic inpainting framework

A novel, automatic restoration framework of large picture areas is proposed in this paper. While the damaged picture areas may have been caused by accident or transmission loss, the challenge consists in repairing the occluded or missing image regions in an undetectable way. In this work, the new inpainting method is used as framework for systematic evaluation of relevant structure extrapolation strategies. The important issue of quality-complexity optimization is thoroughly discussed and efficient inpainting tools are established. Recovery of missing dominant structures is done via tensor voting, regression analysis or a combination of both. Combined structure extrapolation considerably reduces the complexity of structure restoration, compared to genuine tensor voting, with overall graceful to imperceptible degradation of the inpainting result. Furthermore, limitations of the tensor voting approach can be avoided by means of polynomial interpolation with additional savings in terms of complexity.