Intelligent method of texture reconstruction in video sequences based on neural networks
Margarita N. Favorskaya, Mikhail V. Damov, А.Г. Зотин · International Journal of Reasoning-based Intelligent Systems · 2013
We have proposed a novel intelligent method of texture reconstruction which is based on texture estimations of smoothness, structural performance, and isotropy into a missing data region using separated neural networks. Three strategies of boundaries' interpolation of various texture regions by a fast wave algorithm are suggested. The choice of texture reconstruction method is also based on neural network approach by using spatio–temporal parameters of video sequence. The proposed technique was tested for reconstruction of subtitles, damages of information medium, and small objects with area less 8% to 12% of frame area in animation and movies. A background complexity and motions in scenes determine significantly the reconstruction results.