Coding artifact reduction by temporal filtering

Abstract

In this paper the potential improvements by temporal coding artifact reduction are discussed. Whereas conventional spatial methods may suffer from flicker in the processed output due to variable intra frame processing or insufficient removal of coding artifacts, the temporal strategies are capable of flicker reduction and even re-establishment of details missing in the actual input frame. In addition to several temporal filter strategies a new concept for estimating the true motion over multiple input frames is proposed. As a result an image content adaptive spatiotemporal artifact reduction system is presented yielding an even higher output quality.

Spatio-temporal regularization featuring novel temporal priors and multiple reference motion estimation

“…erroneous motion compensation caused by block based coding standards or flicker. In [5] a temporal filtering was presented suitable for this task.…”

“…Moreover, if the level of input information varies over the frames, even some image features not present in the current input frame can be re-established yielding temporally more stable details. A method for multiple reference frame motion estimation suitable for this task was already published in [5]. It can determine the true object motion over several input frames, and fulfils a good compromise between computational complexity and motion vector quality.…”

“…Because motion vector length will increase if it points to a reference frame with a higher distance, length correction is carried out. Further improvements can be achieved by additional predictors, global motion models, internal optical flow update vector generation and hierarchical block splitting to increase the resolution of the motion vector field [5].…”

“…Temporal processing strategies overcome these limitations. A simple but yet efficient temporal filter strategy was presented in [5], mainly benefitting from a highly accurate multi reference frame true motion estimation. In this paper, we describe a novel spatio-temporal and image content adaptive regularization algorithm.…”

“…In addition, we present the incorporation of a novel temporal constraint into the mathematical framework that enables robust processing even in the presence of erroneous motion vectors. Motion information necessary for compensation of temporally neighbored frames is computed by the method described in [5]. Existing spatio-temporal regularization methods optimize have several drawbacks.…”