text_src/14.02__vp8-bitstream__inverse-transforms.txt - webm/bitstream-guide - Git at Google



 #### 14.2 Inverse Transforms                                 {#h-14-02}


 If the Y2 residue block exists (i.e., the macroblock luma mode is not `SPLITMV` or `B_PRED`), it is inverted first (using the inverse WHT) and the element of the result at row i, column j is used as the 0<sup>th</sup> coefficient of the Y subblock at position (i, j), that is, the Y subblock whose index is (i * 4) + j. As discussed in Chapter 13, if the luma mode is `B_PRED` or `SPLITMV`, the 0<sup>th</sup> Y coefficients are part of the residue signal for the subblocks themselves.

 In either case, the inverse transforms for the sixteen Y subblocks and eight chroma subblocks are computed next. All 24 of these inversions are independent of each other; their results may (at least conceptually) be stored in 24 separate 4x4 arrays.

 As is done by the reference decoder, an implementation may wish to represent the prediction and residue buffers as macroblock-sized arrays (that is, a 16x16 Y buffer and two 8x8 chroma buffers). Regarding the inverse DCT implementation given below, this requires a simple adjustment to the address calculation for the resulting residue pixels.


	#### 14.2 Inverse Transforms {#h-14-02}


	If the Y2 residue block exists (i.e., the macroblock luma mode is not `SPLITMV` or `B_PRED`), it is inverted first (using the inverse WHT) and the element of the result at row i, column j is used as the 0<sup>th</sup> coefficient of the Y subblock at position (i, j), that is, the Y subblock whose index is (i * 4) + j. As discussed in Chapter 13, if the luma mode is `B_PRED` or `SPLITMV`, the 0<sup>th</sup> Y coefficients are part of the residue signal for the subblocks themselves.

	In either case, the inverse transforms for the sixteen Y subblocks and eight chroma subblocks are computed next. All 24 of these inversions are independent of each other; their results may (at least conceptually) be stored in 24 separate 4x4 arrays.

	As is done by the reference decoder, an implementation may wish to represent the prediction and residue buffers as macroblock-sized arrays (that is, a 16x16 Y buffer and two 8x8 chroma buffers). Regarding the inverse DCT implementation given below, this requires a simple adjustment to the address calculation for the resulting residue pixels.