Intelligent inpainting system for texture reconstruction in videos with text removal

In this paper we propose an intelligent technique of texture reconstruction in videos which may be also applied in images with some constraints. Traditional approach includes text finding and removal, and texture reconstruction without detailed texture analysis. The novel method extends texture analysis not only in spatial domain, but also in temporal domain of video imagery. This intelligent technique may be used in video editor systems for reconstruction of original videos imagery without artificially imposed small graphic elements (subtitles, clocks, time-codes, signatures and etc.). Our method is mainly directed to localize text regions in video imagery. Following step is a moving estimation in these approximate areas with possible scene changing. Then we classify texture features and apply various reconstruction techniques for missing pixels. Experiments show very good results for isotropic texture reconstruction and good results in the complex cases with distinct texton structures.