Text preserving animation generation using smart device

Abstract

Animation of videos is a modern form of art. The processing of videos of natural scenes is a great way to produce fast and aesthetically pleasing animations. However, the animation process leads to loss of many details in the produced video. In addition, the process of animating a video is time-consuming, especially in resource constrained devices like smart devices. In this paper, we propose a novel framework to animate videos while preserving their text content. A multiplayer simultaneous extensive-form game based on Markov decision process is used to classify the candidate text regions either as text or non-text. A fast animation technique using optimized glass painting is designed to efficiently animate videos with scene text using smart devices. We have used two public datasets, namely, ICDAR 2013 and Hua's video datasets as well as our own dataset created from videos available from Google, to test our framework. The framework is implemented in different smart devices to show its efficiency. © 2017 IEEE.