Frame Rate Up Conversion Algorithm Based On Sparse Representation

At present, several conventional frame up-conversion algorithms based onmotion compensate have been proposed. However, due to the widely application offrame up-conversion algorithm, some new requirements come up, such as the minimalcomputational complexity, easily implementation, meeting the real-time requirement,high quality of reconstruction and so on. And the last one is the fundamentalrequirement, which stimulates the proposal of new frame up-conversion algorithm. Inthe field of computer vision, image super-resolution problem has been successfullysolved by combining sparse representation and dictionary learning, the way torepresenting the frame in the video. Moreover, non-linear regression has been appliedin modeling the relation among data set, the way to describe the relation of frames inthe video. Inspired by these techniques, we propose a new frame up-conversionalgorithm based on sparse representation to achieve better quality of reconstruction.This paper focuses on: How to exploit the dedication of sparse representation anddictionary learning and how to apply these skills in frame up-conversion problem? So,the main works are as follows: sparse representation, dictionary learning, non-linearparameters learning and frame reconstruction. First, to assure the effectiveness ofsparse representation and dictionary learning, verify the ability of representing imagebased on them. Then, learn the dictionary according to former, middle and later frame.To establish a map function among frames, utilize non-linear regression with thesparse representation of each frame. Finally, reconstruct interpolating frame by usingthe dictionary and map function to realize frame up-conversion.The main contributions of this paper are as follows:(1)Combining sparserepresentation and dictionary learning to solve frame up-conversion problem.(2)Weestablish a relationship among former, middle and later frame by using non-linearregression skill in machine learning.(3)We propose a novel frame up-conversionalgorithm based on sparse representation and validate the effectiveness.