Intelligent Texture Reconstruction of Missing Data in Video Sequences Using Neural Networks

The missing data appear in video sequences after removal of non-disabled objects or artifacts. We have proposed an intelligent method of texture reconstruction which novelty consists in a mode of texture estimations using separated neural networks, a boundaries interpolation into a missing data region by a fast wave algorithm, and a texture inpainting considering spatio-temporal parameters of surrounding region. We suggest three strategies of wave algorithm for contour optimization into a missing data region. The proposed technique was tested for visual reconstruction of small missing regions such as subtitles, logotypes and large regions (less 8-12% of frame area). In the first case we have a simplified decision without stage of boundaries approximation, in the second case a background complexity and motions in scene determine significantly the reconstruction results.