Error concealment for 3-D DWT based video codec using iterative thresholding
There exist many applications where video data is compressed and transmitted over unreliable wireless channels, e.g., video transmission for space missions, video delivery from drones, vehicle-to-vehicle video delivery, multimedia wireless sensor networks and so on. In such applications, a video coding algorithm should generate a bit stream, which is robust to packet losses. Video coding standards, such as H.264/AVC and H.265/HEVC, can be utilized for the transmission. Due to the motion estimation and compensation they achieve very high compression efficiency at the expense of a high sensitivity of the video stream to packet loss. As an alternative, video coding based on a 3-D discrete wavelet transform (3-D DWT) encodes each group of wavelet coefficients independently. It provides a good balance between compression efficiency and robustness to packet loss.
In a DWT-based video transmission, an inverse DWT can be performed assuming that all lost coefficients are zero. This approach gives relatively good reconstruction performance, especially if the lowest subbands coefficients are delivered using additional protection, such as duplication of lowest subband coefficients, multiple description coding and/or inter-packet forward error-correction. If neighbor coefficients are delivered, then a lost coefficient can be interpolated utilizing local correlation of a subband coefficient. This recovery method can improve the reconstruction quality without significant increase of computational complexity, but it cannot be efficiently used when many or even all coefficients of a subband are lost.
In this project we consider the error concealment (ERC) problem as a problem of signal recovery from incomplete measurements, which is well-known in image super-resolution and in compressive sensing. We assume that the video sequence has a sparse representation in a known basis different from the DWT, e.g., in a 2-D discrete cosine transform basis. Then we formulate the concealment problem as l1-norm minimization and solve it utilizing an iterative thresholding algorithm. Comparing different thresholding operators we show that Video Block-Matching and 3D filtering (BM3D) provides the best reconstruction by utilizing spatial similarity within a frame and temporal similarity between neighbor frames.
The proposed algorithm has the following advantages:
For more detailed information pleasee see .
2. Performance comparison
In order to provide comparison with ERC via linear interpolation, we applied biorthogonal 4.4 2-D DWT with 3 levels of decomposition for each input frame, then 1-D Haar DWT with 4 levels of decomposition for each group of 16 frames, and SPIHT encoding for each frame. The packetization was simulated in the following way. Each packet contains 11 interleaved coefficients with an interval of 4 for the root subband and all corresponding spatio-temporal descendant coefficients. Then we simulated packet loss using the independent loss model, recovered the video sequence utilizing root subband averaging and futher used the recovered video as initial estimates for the proposed error concealment.
The following pictures demostrate the performance of the proposed algorithm comparing to the root subband interpolation.
The following video demostrates the performance of the proposed algorithm comparing to the root subband interpolation.
Error concealment for SPIHT, Version 1.0 [download]
If you plan to use this software, please also refer to the following paper:
 E.Belyaev, S.Forchhammer, M.Codreanu, Error concealment for 3-D DWT based video codec using iterative thresholding, IEEE Communications Letters, vol.21, Iss.8, pp. 1731-1734, 2017. [download] [draft]