HEVC代码学习6：filterHor和filterVer函数

帧间预测分为运动估计ME和运动补偿MV，其中用到了MV的亚像素搜索，需要使用filterHor和filterVer进行插值。
这里使用的是HM16，在之前版本中分为filterHorLuma、filterHorChroma和filterHorLuma、filterVerChroma。HM16中已经将亮度和色度插值滤波器整合。

我们来从最底层插值滤波器函数filter看起。filter函数代码如下，对同一亚像素位置（如1/4处）处进行插值，插值后的MV存在dst中。

/** 插值滤波器* \brief Apply FIR filter to a block of samples  ** \tparam N          Number of taps  抽头数* \tparam isVertical Flag indicating filtering along vertical direction* \tparam isFirst    Flag indicating whether it is the first filtering operation* \tparam isLast     Flag indicating whether it is the last filtering operation* \param  bitDepth   Bit depth of samples* \param  src        Pointer to source samples  源MV地址* \param  srcStride  Stride of source samples* \param  dst        Pointer to destination samples   插值后的MV地址* \param  dstStride  Stride of destination samples* \param  width      Width of block* \param  height     Height of block* \param  coeff      Pointer to filter taps  抽头系数*/
template<Int N, Bool isVertical, Bool isFirst, Bool isLast>
Void TComInterpolationFilter::filter(Int bitDepth, Pel const *src, Int srcStride, Pel *dst, Int dstStride, Int width, Int height, TFilterCoeff const *coeff)
{Int row, col;Pel c[8];   //系数数组c[0] = coeff[0];   //抽头0系数c[1] = coeff[1];   //抽头1系数if ( N >= 4 )    //N≥4时，取抽头2和3系数（UV分量）{c[2] = coeff[2];c[3] = coeff[3];}if ( N >= 6 )    //N≥6时，取抽头4和5系数{c[4] = coeff[4];c[5] = coeff[5];}if ( N == 8 )    //N=8时，取抽头6和7系数（Y分量）{c[6] = coeff[6];c[7] = coeff[7];}Int cStride = ( isVertical ) ? srcStride : 1;   //垂直情况下步长为srcStride，水平情况下为1src -= ( N/2 - 1 ) * cStride;   //找到前面N/2-1个整数点Int offset;Pel maxVal;Int headRoom = std::max<Int>(2, (IF_INTERNAL_PREC - bitDepth));Int shift    = IF_FILTER_PREC;// with the current settings (IF_INTERNAL_PREC = 14 and IF_FILTER_PREC = 6), though headroom can be// negative for bit depths greater than 14, shift will remain non-negative for bit depths of 8->20assert(shift >= 0);if ( isLast ){shift += (isFirst) ? 0 : headRoom;offset = 1 << (shift - 1);offset += (isFirst) ? 0 : IF_INTERNAL_OFFS << IF_FILTER_PREC;maxVal = (1 << bitDepth) - 1;}else{shift -= (isFirst) ? headRoom : 0;offset = (isFirst) ? -IF_INTERNAL_OFFS << shift : 0;maxVal = 0;}for (row = 0; row < height; row++){for (col = 0; col < width; col++){Int sum;//从N/2-1个整数点开始，取N个整数点乘对应的抽头系数求和。sum  = src[ col + 0 * cStride] * c[0];    sum += src[ col + 1 * cStride] * c[1];    if ( N >= 4 ){sum += src[ col + 2 * cStride] * c[2];sum += src[ col + 3 * cStride] * c[3];}if ( N >= 6 ){sum += src[ col + 4 * cStride] * c[4];sum += src[ col + 5 * cStride] * c[5];}if ( N == 8 ){sum += src[ col + 6 * cStride] * c[6];sum += src[ col + 7 * cStride] * c[7];}Pel val = ( sum + offset ) >> shift;  if ( isLast ){val = ( val < 0 ) ? 0 : val;    //保证不小于0val = ( val > maxVal ) ? maxVal : val;   //保证不越界}dst[col] = val;   //存储插值后的MV}src += srcStride;   dst += dstStride;}
}

再来看上一层的函数，函数代码如下。是一个分类处理，分为普通操作、第一次操作和最后一次操作，分别传入不同参数调用底层filter函数进行插值。这里直接通过N来区分是亮度插值和色度插值。

/*** \brief Filter a block of samples (horizontal)** \tparam N          Number of taps 抽头数：Y-8，UV-4* \param  bitDepth   Bit depth of samples* \param  src        Pointer to source samples* \param  srcStride  Stride of source samples* \param  dst        Pointer to destination samples* \param  dstStride  Stride of destination samples* \param  width      Width of block* \param  height     Height of block* \param  isLast     Flag indicating whether it is the last filtering operation* \param  coeff      Pointer to filter taps 抽头系数*/
template<Int N>     //插值
Void TComInterpolationFilter::filterHor(Int bitDepth, Pel *src, Int srcStride, Pel *dst, Int dstStride, Int width, Int height, Bool isLast, TFilterCoeff const *coeff)
{if ( isLast ){filter<N, false, true, true>(bitDepth, src, srcStride, dst, dstStride, width, height, coeff);}else{filter<N, false, true, false>(bitDepth, src, srcStride, dst, dstStride, width, height, coeff);}
}template<Int N>
Void TComInterpolationFilter::filterVer(Int bitDepth, Pel *src, Int srcStride, Pel *dst, Int dstStride, Int width, Int height, Bool isFirst, Bool isLast, TFilterCoeff const *coeff)
{if ( isFirst && isLast ){filter<N, true, true, true>(bitDepth, src, srcStride, dst, dstStride, width, height, coeff);}else if ( isFirst && !isLast ){filter<N, true, true, false>(bitDepth, src, srcStride, dst, dstStride, width, height, coeff);}else if ( !isFirst && isLast ){filter<N, true, false, true>(bitDepth, src, srcStride, dst, dstStride, width, height, coeff);}else{filter<N, true, false, false>(bitDepth, src, srcStride, dst, dstStride, width, height, coeff);}
}

最后来看最上层的filterHor和filterVer，代码如下，也是一个分类处理，调用下层函数处理。分了三种情况：frac = 0、亮度和色度。frac = 0，即整数，不需要插值。

/*** \brief Filter a block of Luma/Chroma samples (horizontal)    ** \param  compID     Chroma component ID* \param  src        Pointer to source samples* \param  srcStride  Stride of source samples* \param  dst        Pointer to destination samples* \param  dstStride  Stride of destination samples* \param  width      Width of block* \param  height     Height of block* \param  frac       Fractional sample offset  分数偏移* \param  isLast     Flag indicating whether it is the last filtering operation* \param  fmt        Chroma format* \param  bitDepth   Bit depth*/
Void TComInterpolationFilter::filterHor(const ComponentID compID, Pel *src, Int srcStride, Pel *dst, Int dstStride, Int width, Int height, Int frac, Bool isLast, const ChromaFormat fmt, const Int bitDepth )
{if ( frac == 0 )   //整数直接复制{filterCopy(bitDepth, src, srcStride, dst, dstStride, width, height, true, isLast );}else if (isLuma(compID))   //Y分量小数插值{assert(frac >= 0 && frac < LUMA_INTERPOLATION_FILTER_SUB_SAMPLE_POSITIONS);filterHor<NTAPS_LUMA>(bitDepth, src, srcStride, dst, dstStride, width, height, isLast, m_lumaFilter[frac]);}else{const UInt csx = getComponentScaleX(compID, fmt);assert(frac >=0 && csx<2 && (frac<<(1-csx)) < CHROMA_INTERPOLATION_FILTER_SUB_SAMPLE_POSITIONS);filterHor<NTAPS_CHROMA>(bitDepth, src, srcStride, dst, dstStride, width, height, isLast, m_chromaFilter[frac<<(1-csx)]);}
}Void TComInterpolationFilter::filterVer(const ComponentID compID, Pel *src, Int srcStride, Pel *dst, Int dstStride, Int width, Int height, Int frac, Bool isFirst, Bool isLast, const ChromaFormat fmt, const Int bitDepth )
{if ( frac == 0 ){filterCopy(bitDepth, src, srcStride, dst, dstStride, width, height, isFirst, isLast );}else if (isLuma(compID)){assert(frac >= 0 && frac < LUMA_INTERPOLATION_FILTER_SUB_SAMPLE_POSITIONS);filterVer<NTAPS_LUMA>(bitDepth, src, srcStride, dst, dstStride, width, height, isFirst, isLast, m_lumaFilter[frac]);}else{const UInt csy = getComponentScaleY(compID, fmt);assert(frac >=0 && csy<2 && (frac<<(1-csy)) < CHROMA_INTERPOLATION_FILTER_SUB_SAMPLE_POSITIONS);filterVer<NTAPS_CHROMA>(bitDepth, src, srcStride, dst, dstStride, width, height, isFirst, isLast, m_chromaFilter[frac<<(1-csy)]);}
}

HEVC代码学习6：filterHor和filterVer函数相关推荐

HEVC代码学习42：estIntraPredLumaQT函数
在之前的 HEVC代码学习37:帧内预测代码整体学习中已经提到,estIntraPredLumaQT是亮度帧内预测的入口函数,下面将对该函数进行详细学习. estIntraPredLumaQT中完成 ...
HEVC代码学习39：decodeCtu和xDecodeCU函数
在之前 HEVC代码学习38:decompressSlice函数学习中提到,解码slice会遍历所有CTU,调用decodeCtu和decompressCtu解码每一个CTU.下面就来学习一下dec ...
HEVC代码学习——帧间预测：预测MV获取（xEstimateMvPredAMVP、fillMVPCand）
HEVC帧间预测在AMVP模式下是依靠xEstimateMvPredAMVP函数获取预测MV(MVP)的. 这部分内容的学习还可以参考这两篇博客: HEVC代码学习15:AMVP相关函数 HM编码器代 ...
HEVC代码学习：帧间预测——MVP过程中MV的获取、传递及存储
作为一个视频编码小白,最近开始着手啃HEVC帧间预测的代码,想用博客记录一下自己的学习过程,也想与大家分享.交流一下. HEVC代码的学习主要是参考两位大神岳麓吹雪.NB_vol_1的博客以及HM参考 ...
H.266代码学习：decodeCtu和xDecodeCU函数
之前 HEVC代码学习39:decodeCtu和xDecodeCU函数中对HM中的CTU解码函数进行了学习,这里来学习一下JEM中的. 首先需要强调的是,这里只是用来解码flag.系数等,没有进行预 ...
H.266代码学习：decompressCtu和xDecompressCU函数
今天来学习一下JEM的decompressCtu和xDecompressCU函数.之前在 H.266代码学习:decodeCtu和xDecodeCU函数学习了的学习中提到,decodeCtu和xDe ...
HEVC代码学习15：AMVP相关函数
在HEVC中,使用了AMVP技术,利用空域和时域上的运动向量的相关性,为当前PU建立候选预测MV(MVP)列表.编码器从中选出最优的预测MV,并对MV进行差分编码:解码端会构造相同的列表,仅需要运动向 ...
H.266/VVC-VTM代码学习27-VTM中编码器主函数逻辑
H.266/VVC专栏传送上一篇:H.266/VVC-VTM代码学习26-VTM中RDcost的计算与λ的设定(二) 下一篇:持续创作中- 目录 H.266/VVC专栏传送前言一.简介二.代码 ...
HEVC代码学习13：predInterSearch函数
在上一章的xCheckRDCostInter学习中,我们知道了,进行帧间搜索的入口实际是predInterSearch,今天我们就来对他进行学习. 推荐看大神博客 http://blog.csdn.n ...

HEVC代码学习6：filterHor和filterVer函数

HEVC代码学习6：filterHor和filterVer函数相关推荐

最新文章

热门文章