其他分享
首页 > 其他分享> > opencv 的pyramid down函数verilog实现

opencv 的pyramid down函数verilog实现

作者:互联网

实现opencv如下的函数,在pyramid lk算法中用到,其他地方也有用得到,单独拿出来。

代码共享在github:   https://github.com/tishi43/pyr_down


template<class CastOp, class VecOp> void

pyrDown_( const Mat& _src, Mat& _dst, int borderType )

{

    const int PD_SZ = 5;

    typedef typename CastOp::type1 WT;

    typedef typename CastOp::rtype T;



    CV_Assert( !_src.empty() );

    Size ssize = _src.size(), dsize = _dst.size();

    int cn = _src.channels();

    int bufstep = (int)alignSize(dsize.width*cn, 16);

    AutoBuffer<WT> _buf(bufstep*PD_SZ + 16);

    WT* buf = alignPtr((WT*)_buf, 16);

    int tabL[CV_CN_MAX*(PD_SZ+2)], tabR[CV_CN_MAX*(PD_SZ+2)];

    AutoBuffer<int> _tabM(dsize.width*cn);

    int* tabM = _tabM;

    WT* rows[PD_SZ];

    CastOp castOp;

    VecOp vecOp;

char path[64] = { 0 };

int pyramid_level = 0;

sprintf(path, "hor%d.txt", pyramid_level);

fopen_s(&f_hor, path, "wb");

sprintf(path, "ver%d.txt", pyramid_level);

fopen_s(&f_ver, "ver.txt", "wb");



    CV_Assert( ssize.width > 0 && ssize.height > 0 &&

               std::abs(dsize.width*2 - ssize.width) <= 2 &&

               std::abs(dsize.height*2 - ssize.height) <= 2 );

    int k, x, sy0 = -PD_SZ/2, sy = sy0, width0 = std::min((ssize.width-PD_SZ/2-1)/2 + 1, dsize.width);



    for( x = 0; x <= PD_SZ+1; x++ )

    {

        int sx0 = borderInterpolate(x - PD_SZ/2, ssize.width, borderType)*cn;

        int sx1 = borderInterpolate(x + width0*2 - PD_SZ/2, ssize.width, borderType)*cn;

        for( k = 0; k < cn; k++ )

        {

            tabL[x*cn + k] = sx0 + k;

            tabR[x*cn + k] = sx1 + k;

        }

    }



    ssize.width *= cn;

    dsize.width *= cn;

    width0 *= cn;



    for( x = 0; x < dsize.width; x++ )

        tabM[x] = (x/cn)*2*cn + x % cn;



    for( int y = 0; y < dsize.height; y++ )

    {

        T* dst = _dst.ptr<T>(y);

        WT *row0, *row1, *row2, *row3, *row4;



        // fill the ring buffer (horizontal convolution and decimation)

        for( ; sy <= y*2 + 2; sy++ )

        {

            WT* row = buf + ((sy - sy0) % PD_SZ)*bufstep;

            int _sy = borderInterpolate(sy, ssize.height, borderType);

            const T* src = _src.ptr<T>(_sy);

            int limit = cn;

            const int* tab = tabL;



            for( x = 0;;)

            {

                for( ; x < limit; x++ )

                {

                    row[x] = src[tab[x+cn*2]]*6 + (src[tab[x+cn]] + src[tab[x+cn*3]])*4 +

                        src[tab[x]] + src[tab[x+cn*4]];

if (sy >= 0)

Log_hor("y=%d x=%d conv=%d\n",sy,x,row[x]);

                }



                if( x == dsize.width )

                    break;



                if( cn == 1 )

                {

for (; x < width0; x++){

row[x] = src[x * 2] * 6 + (src[x * 2 - 1] + src[x * 2 + 1]) * 4 +

src[x * 2 - 2] + src[x * 2 + 2];

if (sy >= 0)

Log_hor("y=%d x=%d conv=%d\n",sy, x, row[x]);

}



                }

                else if( cn == 3 )

                {

                    for( ; x < width0; x += 3 )

                    {

                        const T* s = src + x*2;

                        WT t0 = s[0]*6 + (s[-3] + s[3])*4 + s[-6] + s[6];

                        WT t1 = s[1]*6 + (s[-2] + s[4])*4 + s[-5] + s[7];

                        WT t2 = s[2]*6 + (s[-1] + s[5])*4 + s[-4] + s[8];

                        row[x] = t0; row[x+1] = t1; row[x+2] = t2;

                    }

                }

                else if( cn == 4 )

                {

                    for( ; x < width0; x += 4 )

                    {

                        const T* s = src + x*2;

                        WT t0 = s[0]*6 + (s[-4] + s[4])*4 + s[-8] + s[8];

                        WT t1 = s[1]*6 + (s[-3] + s[5])*4 + s[-7] + s[9];

                        row[x] = t0; row[x+1] = t1;

                        t0 = s[2]*6 + (s[-2] + s[6])*4 + s[-6] + s[10];

                        t1 = s[3]*6 + (s[-1] + s[7])*4 + s[-5] + s[11];

                        row[x+2] = t0; row[x+3] = t1;

                    }

                }

                else

                {

                    for( ; x < width0; x++ )

                    {

                        int sx = tabM[x];

                        row[x] = src[sx]*6 + (src[sx - cn] + src[sx + cn])*4 +

                            src[sx - cn*2] + src[sx + cn*2];

                    }

                }



                limit = dsize.width;

                tab = tabR - x;

            }

        }



        // do vertical convolution and decimation and write the result to the destination image

        for( k = 0; k < PD_SZ; k++ )

            rows[k] = buf + ((y*2 - PD_SZ/2 + k - sy0) % PD_SZ)*bufstep;

        row0 = rows[0]; row1 = rows[1]; row2 = rows[2]; row3 = rows[3]; row4 = rows[4];



        //x = vecOp(rows, dst, (int)_dst.step, dsize.width);

x = 0;

for (; x < dsize.width; x++){

dst[x] = castOp(row2[x] * 6 + (row1[x] + row3[x]) * 4 + row0[x] + row4[x]);

Log_ver("y=%d x=%d conv=%d cast=%d(0x%x)\n", y, x, row2[x] * 6 + (row1[x] + row3[x]) * 4 + row0[x] + row4[x]+128, dst[x], dst[x]);

}

            

    }



fclose(f_hor);

fclose(f_ver);

}

标签:src,pyramid,cn,width,int,down,opencv,WT,row
来源: https://blog.csdn.net/tianqishi/article/details/120285753