Dominique Thoreau

Fast encoding algorithms for geometry-adaptive block partitioning

State-of-the-art video compression technologies, such as MPEG-4 AVC/H.264 or the new HEVC standard being developed by ISO MPEG and ITU-T VCEG, make use of tree-structured block partitioning for motion compensation. Such motion partitioning only captures horizontal and vertical motion boundaries. To better match actual motion frontiers, Geometry adaptive block partitioning (GEO) has been explored for several years. GEO enables splitting a block using non-horizontal or non-vertical line. Although noticeable coding efficiency gains can be obtained, GEO involves a significant increase of the number of modes to be tested with therefore a high impact on encoding complexity. This paper presents fast algorithms aiming at controlling the complexity while saving the coding efficiency gains of GEO. Experimental results are provided on top of the HEVC standard, demonstrating first the efficiency of the GEO tool. Simplified versions, offering noticeable complexity reduction with limited rate-distortion performance loss, are also demonstrated and compared.

Sparse shift-DCT spatial prediction

In this paper, we propose a new intra prediction scheme based on hard thresholding applied on shift-DCT coefficients. The explored intra prediction method, validated in an H.264/AVC framework, aims at better predicting complex 2D patterns that cannot be properly extrapolated using H.264/AVC directional intra prediction modes. The basic explored algorithm makes use of neighbor pixels of the current block to predict the current signal using a shift-DCT based extrapolation. Different versions of the algorithm, corresponding to different hard thresholding parameter derivation processes, are described. Experimental results show that the proposed intra coding prediction method can significantly improve the coding efficiency compared to the directional predictive modes of H.264/AVC, with an average BD-rate gain of 6.2% observed on a various pictures test set.

Sparse representations for spatial prediction and texture refinement

In this work, we propose a novel approach for signal prediction based on the use of sparse signal representations and Matching Pursuit (MP) techniques. The paper first focuses on spatial texture prediction in a conventional block-based hybrid coding scheme and secondly addresses inter-layer prediction in a scalable video coding (SVC) framework. For spatial prediction the signal reconstruction of the block to predict is based on basis functions selected with the MP iterative algorithm, to best match a causal neighborhood. Inter-layer MP based prediction employs base layer upsampled components additionally to the causal neighborhood in order to improve the representation of high frequencies. New solutions are proposed for efficiently deriving and exploiting the atoms dictionary through phase refinement and mono-dimensional basis functions. Experimental results indicate noticeable improvement of rate/distortion performance compared to the standard prediction methods as specified in H.264/AVC and its extension SVC.