Table 1 Enhancements to PDWZ
Module | Summary | References |
|---|---|---|
Rate-control | Encoder rate-control algorithm without feedback channel at the expense of increased encoder complexity An efficient block motion-estimation algorithm at encoder for estimating bitplane error probability and low complexity side information A code mode decision algorithm at encoder to improve coding performance | (Du and Shen 2009) |
Encoder based rate-allocation algorithm that computes the number of bits to encode each WZ frame without significantly increasing encoder complexity Uses a Laplacian random variable to represent the difference in bitplane values between the original frame and corresponding side information Defines a probability mass function to estimate the aforementioned random variable Estimates the bit error probability for each bitplane based on the error correcting capacity of the turbo code and frame rate of the video Prevents increase in distortion due to excessive errors in decoded bitplanes by discarding parity bits and sets decoded frame to side information if residual error probability estimated at decoder is above a given threshold | (Morbee et al. 2007) | |
Decoding algorithm | A mode decision scheme that can be executed at encoder or decoder (or both) to determine if the correlation noise estimation between a frame to be encoded and its side information is weak, and if block-based intra-frame coding should be selected instead of block-based WZ coding Shows that the relationship between the frame to be encoded and the side information at the decoder (defined as correlation noise statistics) is not spatially stationary Determines the selection criteria for mode decision by exploiting spatial and temporal statistics Creates a binary map whose entries indicate which blocks of a frame to be encoded should be intra- or WZ-coded. A simple entropy coding algorithm is used for efficient processing of this information | (Tagliasacchi et al. 2006a) |
A coding distortion model that can be used to determine the value of coding parameters such as quantization step size, target distortion, distortion predictions under certain coding constraints Selects the quantization step size of each video frame to meet the target distortion Shows that the accuracy of distortion predictions is limited by computation capacity of PDWZ encoders and the stability of distortion constraints | (Roca et al. 2007) | |
A decoding algorithm based on turbo codes that requires a small subset of parity bits at the decoder for each WZ frame, and exploits the temporal correlation of the video sequence using previously reconstructed frame as noisy side information Mismatches between the side information and frame to be decoded are represented by pixel values and parity bits Uses a suboptimal approach to convert pixel values to soft information for the parity bits Uses hyper-trellis codes to combine multiple states of original trellis code Improves codec performance without increasing decoder’s complexity | (Avudainayagam et al. 2008) | |
A decoding algorithm that incorporates side information computed from either lossless or quantized frames A mode decision scheme at the decoder (similar to the one in Fan et al. 2010) is used to improve the correlation noise statistics | (Trapanese et al. 2005a) | |
Side information | An extrapolation module to generate side information based on motion field smoothening filter to enhance performance of a low-delay PDWZ codec Uses overlapped motion estimation, in particular motion field smoothening filtering and spatial-interpolation for un-overlapped regions | (Natário et al. 2005) |
Encoder based motion-compensation module that sends hash codewords of the frame to be decoded to the decoder Hash codewords complements the side information and lead to efficient frame reconstruction Enables low-complexity encoding while maintaining high compression ratio | (Aaron et al. 2004a) | |
Correlation noise modeling | A correlation noise estimation module that performs online estimation of the error distribution at the decoder A temporal model that estimates correlation between frames under different levels of granularity such as: frame, block and pixel levels An improved rate-distortion performance at lower granularity level Collects correlation noise statistics locally on a block-by-block basis and at pixel level | (Brites et al. 2006) |
An enhanced correlation noise model with reasonable coding efficiency gain Shows that a Laplacian model is not an optimal choice to represent the distribution of correlation noise, since the rate at which the tails of the model decreases to zero is slower than the empirical distribution (see Figure 3 of Macchiavello et al. 2007) Presents improved modeling of the tails for the turbo decoding process A turbo decoder that assigns a higher likelihood to estimated values far apart from the corresponding side information to increase the chance of decoding outliers and enhance the reconstruction quality | (Trapanese et al. 2005b) |