HEVC代码追踪（十一。五）:运动估计/补偿之xTZSearch

Void TEncSearch::xTZSearch( TComDataCU* pcCU, TComPattern* pcPatternKey, Pel* piRefY, Int iRefStride, TComMv* pcMvSrchRngLT, TComMv* pcMvSrchRngRB, TComMv& rcMv, UInt& ruiSAD )
{//!< 确定运动估计搜索范围的边界
  Int   iSrchRngHorLeft   = pcMvSrchRngLT->getHor();
  Int   iSrchRngHorRight  = pcMvSrchRngRB->getHor();
  Int   iSrchRngVerTop    = pcMvSrchRngLT->getVer();
  Int   iSrchRngVerBottom = pcMvSrchRngRB->getVer();
  //!< 以宏定义方式对TZSearch的相关参数进行设置
  TZ_SEARCH_CONFIGURATION

  UInt uiSearchRange = m_iSearchRange;
  pcCU->clipMv( rcMv );
  rcMv >>= 2;
  // init TZSearchStruct
  IntTZSearchStruct cStruct;
  cStruct.iYStride    = iRefStride;
  cStruct.piRefY      = piRefY;
  cStruct.uiBestSad   = MAX_UINT;

  // set rcMv (Median predictor) as start point and as best point
  xTZSearchHelp( pcPatternKey, cStruct, rcMv.getHor(), rcMv.getVer(), 0, 0 );//!< 中值预测

  // test whether one of PRED_A, PRED_B, PRED_C MV is better start point than Median predictor
  if ( bTestOtherPredictedMV )
  {
    for ( UInt index = 0; index < 3; index++ )
    {
      TComMv cMv = m_acMvPredictors[index];
      pcCU->clipMv( cMv );
      cMv >>= 2;
      xTZSearchHelp( pcPatternKey, cStruct, cMv.getHor(), cMv.getVer(), 0, 0 ); //!< A, B, C相邻PU的mv
    }
  }

  // test whether zero Mv is better start point than Median predictor
  if ( bTestZeroVector )
  {
    xTZSearchHelp( pcPatternKey, cStruct, 0, 0, 0, 0 );//!< 零mv
  }

  // start search，从以前面几个mv作为搜索起点得到的最好的位置开始进行接下来的搜索
  Int  iDist = 0;
  Int  iStartX = cStruct.iBestX;
  Int  iStartY = cStruct.iBestY;

  // first search
  for ( iDist = 1; iDist <= (Int)uiSearchRange; iDist*=2 )//!< 以2的幂次逐步扩大搜索步长
  {
    if ( bFirstSearchDiamond == 1 )
    {
      xTZ8PointDiamondSearch ( pcPatternKey, cStruct, pcMvSrchRngLT, pcMvSrchRngRB, iStartX, iStartY, iDist );
    }
    else
    {
      xTZ8PointSquareSearch  ( pcPatternKey, cStruct, pcMvSrchRngLT, pcMvSrchRngRB, iStartX, iStartY, iDist );
    }

    if ( bFirstSearchStop && ( cStruct.uiBestRound >= uiFirstSearchRounds ) ) // stop criterion
    {
      break;
    }
  }

  // test whether zero Mv is a better start point than Median predictor
  if ( bTestZeroVectorStart && ((cStruct.iBestX != 0) || (cStruct.iBestY != 0)) )
  {
    xTZSearchHelp( pcPatternKey, cStruct, 0, 0, 0, 0 );
    if ( (cStruct.iBestX == 0) && (cStruct.iBestY == 0) )
    {
      // test its neighborhood
      for ( iDist = 1; iDist <= (Int)uiSearchRange; iDist*=2 )
      {
        xTZ8PointDiamondSearch( pcPatternKey, cStruct, pcMvSrchRngLT, pcMvSrchRngRB, 0, 0, iDist );
        if ( bTestZeroVectorStop && (cStruct.uiBestRound > 0) ) // stop criterion
        {
          break;
        }
      }
    }
  }

  // calculate only 2 missing points instead 8 points if cStruct.uiBestDistance == 1
  if ( cStruct.uiBestDistance == 1 )//!< 当最佳搜索步长等于1时，补充搜索前面8点钻石扫描遗漏的两点
  {
    cStruct.uiBestDistance = 0;
    xTZ2PointSearch( pcPatternKey, cStruct, pcMvSrchRngLT, pcMvSrchRngRB );
  }

  // raster search if distance is too big
  if ( bEnableRasterSearch && ( ((Int)(cStruct.uiBestDistance) > iRaster) || bAlwaysRasterSearch ) )//!< bEnableRasterSearch default is 1, iRaster default is 5
  {//!< 当前面搜索得到的最佳步长过大时，改用光栅搜索法，步长定为iRaster，搜索范围为设定的运动估计范围
    cStruct.uiBestDistance = iRaster;
    for ( iStartY = iSrchRngVerTop; iStartY <= iSrchRngVerBottom; iStartY += iRaster )
    {
      for ( iStartX = iSrchRngHorLeft; iStartX <= iSrchRngHorRight; iStartX += iRaster )
      {
        xTZSearchHelp( pcPatternKey, cStruct, iStartX, iStartY, 0, iRaster );
      }
    }
  }

  // raster refinement
  if ( bRasterRefinementEnable && cStruct.uiBestDistance > 0 )
  {
    while ( cStruct.uiBestDistance > 0 )
    {
      iStartX = cStruct.iBestX;
      iStartY = cStruct.iBestY;
      if ( cStruct.uiBestDistance > 1 )
      {
        iDist = cStruct.uiBestDistance >>= 1;
        if ( bRasterRefinementDiamond == 1 )
        {
          xTZ8PointDiamondSearch ( pcPatternKey, cStruct, pcMvSrchRngLT, pcMvSrchRngRB, iStartX, iStartY, iDist );
        }
        else
        {
          xTZ8PointSquareSearch  ( pcPatternKey, cStruct, pcMvSrchRngLT, pcMvSrchRngRB, iStartX, iStartY, iDist );
        }
      }

      // calculate only 2 missing points instead 8 points if cStruct.uiBestDistance == 1
      if ( cStruct.uiBestDistance == 1 )
      {
        cStruct.uiBestDistance = 0;
        if ( cStruct.ucPointNr != 0 )
        {
          xTZ2PointSearch( pcPatternKey, cStruct, pcMvSrchRngLT, pcMvSrchRngRB );
        }
      }
    }
  }

  // start refinement
  if ( bStarRefinementEnable && cStruct.uiBestDistance > 0 )
  {
    while ( cStruct.uiBestDistance > 0 )
    {//!< 在经过了上面几个步骤的搜索后，从最佳点开始进行第2次的8点钻石扫描以及利用两点扫描对遗漏点进行补充
      iStartX = cStruct.iBestX;
      iStartY = cStruct.iBestY;
      cStruct.uiBestDistance = 0;
      cStruct.ucPointNr = 0;
      for ( iDist = 1; iDist < (Int)uiSearchRange + 1; iDist*=2 )
      {
        if ( bStarRefinementDiamond == 1 )
        {
          xTZ8PointDiamondSearch ( pcPatternKey, cStruct, pcMvSrchRngLT, pcMvSrchRngRB, iStartX, iStartY, iDist );
        }
        else
        {
          xTZ8PointSquareSearch  ( pcPatternKey, cStruct, pcMvSrchRngLT, pcMvSrchRngRB, iStartX, iStartY, iDist );
        }
        if ( bStarRefinementStop && (cStruct.uiBestRound >= uiStarRefinementRounds) ) // stop criterion
        {
          break;
        }
      }

      // calculate only 2 missing points instead 8 points if cStrukt.uiBestDistance == 1
      if ( cStruct.uiBestDistance == 1 )
      {
        cStruct.uiBestDistance = 0;
        if ( cStruct.ucPointNr != 0 )
        {
          xTZ2PointSearch( pcPatternKey, cStruct, pcMvSrchRngLT, pcMvSrchRngRB );
        }
      }
    }
  }

  // write out best match，获得最佳匹配结果，mv和SAD
  rcMv.set( cStruct.iBestX, cStruct.iBestY );
  ruiSAD = cStruct.uiBestSad - m_pcRdCost->getCost( cStruct.iBestX, cStruct.iBestY );
}

TZSearch的基本流程：

1、搜索预测得到的mv所指向的点：中值预测mv，当前PU的左，上及右上PU的mv，还有零运动矢量（0,0）；

2、在步骤1中找到匹配误差最小的点作为接下来搜索的起始点；

3、步长从1开始，以2的指数递增，进行8点钻石搜索，该步骤中可以设置搜索的最大次数（以某个步长遍历一遍就算1次）；

4、如果步骤3搜索得到的最佳步长为1，则需要以该最佳点为起点做1次两点钻石搜索，因为前面8点搜索的时候，这个最佳点的8个邻点会有两个没有搜索到；

5、如果步骤3搜索得到的最佳步长大于某个阈值（iRaster），则以步骤2得到的点作为起点，做步长为iRaster的光栅扫描（即在运动搜索的范围内遍历所有点）；

6、 最后，在经过前面1~5歩之后，以得到的最佳点为起点，再次重复步骤3和4；

7、保存最佳mv和SAD。