cvderiv.cpp

opencv库在TI DM6437上的移植,目前包括两个库cv.lib和cxcore.lib的工程

                for( i = 0; i <= width - 2; i += 2 )
                {
                    int s0 = src0[i] - src1[i]*2 + src2[i] +
                             src0[i+width] + src1[i+width]*2 + src2[i+width];
                    int s1 = src0[i+1] - src1[i+1]*2 + src2[i+1] +
                             src0[i+width+1] + src1[i+width+1]*2 + src2[i+width+1];
                    dst[i] = (short)s0; dst[i+1] = (short)s1;
                }
            else
                for( i = 0; i <= width - 2; i += 2 )
                {
                    int s0 = CV_DESCALE(src0[i] - src1[i]*2 + src2[i] +
                             src0[i+width] + src1[i+width]*2 + src2[i+width], 2);
                    int s1 = CV_DESCALE(src0[i+1] - src1[i+1]*2 + src2[i+1] +
                             src0[i+width+1] + src1[i+width+1]*2 + src2[i+width+1],2);
                    dst[i] = (short)s0; dst[i+1] = (short)s1;
                }
        }
        else if( ksize == 5 )
        {
            const int *src0 = src[-2], *src1 = src[-1], *src2 = src[0], *src3 = src[1], *src4 = src[2];

            if( !normalized )
                for( i = 0; i <= width - 2; i += 2 )
                {
                    int s0 = src0[i] - src2[i]*2 + src4[i] + src0[i+width] + src4[i+width] +
                             (src1[i+width] + src3[i+width])*4 + src2[i+width]*6;
                    int s1 = src0[i+1] - src2[i+1]*2 + src4[i+1] + src0[i+width+1] +
                             src4[i+width+1] + (src1[i+width+1] + src3[i+width+1])*4 +
                             src2[i+width+1]*6;
                    dst[i] = (short)s0; dst[i+1] = (short)s1;
                }
            else
                for( i = 0; i <= width - 2; i += 2 )
                {
                    int s0 = CV_DESCALE(src0[i] - src2[i]*2 + src4[i] +
                             src0[i+width] + src4[i+width] +
                             (src1[i+width] + src3[i+width])*4 + src2[i+width]*6, 4);
                    int s1 = CV_DESCALE(src0[i+1] - src2[i+1]*2 + src4[i+1] +
                             src0[i+width+1] + src4[i+width+1] +
                             (src1[i+width+1] + src3[i+width+1])*4 + src2[i+width+1]*6, 4);
                    dst[i] = (short)s0; dst[i+1] = (short)s1;
                }
        }
        else
        {
            if( !normalized )
                for( i = 0; i <= width - 2; i += 2 )
                {
                    int s0 = kx[0]*src[0][i] + ky[0]*src[0][i+width];
                    int s1 = kx[0]*src[0][i+1] + ky[0]*src[0][i+width+1];

                    for( k = 1; k <= ksize2; k++ )
                    {
                        const int* src1 = src[k] + i, *src2 = src[-k] + i;
                        int fx = kx[k], fy = ky[k];
                        s0 += fx*(src1[0] + src2[0]) + fy*(src1[width] + src2[width]);
                        s1 += fx*(src1[1] + src2[1]) + fy*(src1[width+1] + src2[width+1]);
                    }

                    dst[i] = CV_CAST_16S(s0); dst[i+1] = CV_CAST_16S(s1);
                }
            else
                for( i = 0; i <= width - 2; i += 2 )
                {
                    int s0 = kx[0]*src[0][i] + ky[0]*src[0][i+width];
                    int s1 = kx[0]*src[0][i+1] + ky[0]*src[0][i+width+1];

                    for( k = 1; k <= ksize2; k++ )
                    {
                        const int* src1 = src[k] + i, *src2 = src[-k] + i;
                        int fx = kx[k], fy = ky[k];
                        s0 += fx*(src1[0] + src2[0]) + fy*(src1[width] + src2[width]);
                        s1 += fx*(src1[1] + src2[1]) + fy*(src1[width+1] + src2[width+1]);
                    }

                    s0 = CV_DESCALE( s0, shift ); s1 = CV_DESCALE( s1, shift );
                    dst[i] = (short)s0; dst[i+1] = (short)s1;
                }
        }

        for( ; i < width; i++ )
        {
            int s0 = kx[0]*src[0][i] + ky[0]*src[0][i+width];
            for( k = 1; k <= ksize2; k++ )
            {
                const int* src1 = src[k] + i, *src2 = src[-k] + i;
                s0 += kx[k]*(src1[0] + src2[0]) + ky[k]*(src1[width] + src2[width]);
            }
            s0 = (s0 + delta) >> shift;
            dst[i] = CV_CAST_16S(s0);
        }
    }
}


static void
icvLaplaceCol_32f( const float** src, float* dst,
                   int dst_step, int count, void* params )
{
    const CvLaplaceFilter* state = (const CvLaplaceFilter*)params;
    const CvMat* _kx = state->get_x_kernel();
    const CvMat* _ky = state->get_y_kernel();
    const float* kx = (const float*)_kx->data.ptr;
    const float* ky = (const float*)_ky->data.ptr;
    int ksize = _kx->cols + _kx->rows - 1, ksize2 = ksize/2;
    int i = 0, k, width = state->get_width();
    int cn = CV_MAT_CN(state->get_src_type());
    bool basic_laplacian = state->is_basic_laplacian();
    bool normalized = state->is_normalized();
    float scale = 1.f/(1 << (ksize - 1));
    
    width *= cn;
    src += ksize2;
    kx += ksize2;
    ky += ksize2;
    dst_step /= sizeof(dst[0]);

    if( basic_laplacian || !normalized )
    {
        normalized = false;
        scale = 1.f;
    }

    for( ; count--; dst += dst_step, src++ )
    {
        if( ksize == 3 )
        {
            const float *src0 = src[-1], *src1 = src[0], *src2 = src[1];
            if( basic_laplacian )
            {
                for( i = 0; i <= width - 2; i += 2 )
                {
                    float s0 = src0[i] - src1[i]*2 + src2[i] + src1[i+width];
                    float s1 = src0[i+1] - src1[i+1]*2 + src2[i+1] + src1[i+width+1];
                    dst[i] = s0; dst[i+1] = s1;
                }

                for( ; i < width; i++ )
                    dst[i] = src0[i] - src1[i]*2 + src2[i] + src1[i+width];
            }
            else if( !normalized )
                for( i = 0; i <= width - 2; i += 2 )
                {
                    float s0 = src0[i] - src1[i]*2 + src2[i] +
                             src0[i+width] + src1[i+width]*2 + src2[i+width];
                    float s1 = src0[i+1] - src1[i+1]*2 + src2[i+1] +
                             src0[i+width+1] + src1[i+width+1]*2 + src2[i+width+1];
                    dst[i] = s0; dst[i+1] = s1;
                }
            else
                for( i = 0; i <= width - 2; i += 2 )
                {
                    float s0 = (src0[i] - src1[i]*2 + src2[i] +
                             src0[i+width] + src1[i+width]*2 + src2[i+width])*scale;
                    float s1 = (src0[i+1] - src1[i+1]*2 + src2[i+1] +
                             src0[i+width+1] + src1[i+width+1]*2 + src2[i+width+1])*scale;
                    dst[i] = s0; dst[i+1] = s1;
                }
        }
        else if( ksize == 5 )
        {
            const float *src0 = src[-2], *src1 = src[-1], *src2 = src[0], *src3 = src[1], *src4 = src[2];
            for( i = 0; i <= width - 2; i += 2 )
            {
                float s0 = (src0[i] - src2[i]*2 + src4[i] +
                         src0[i+width] + src4[i+width] +
                         (src1[i+width] + src3[i+width])*4 + src2[i+width]*6)*scale;
                float s1 = (src0[i+1] - src2[i+1]*2 + src4[i+1] +
                         src0[i+width+1] + src4[i+width+1] +
                         (src1[i+width+1] + src3[i+width+1])*4 + src2[i+width+1]*6)*scale;
                dst[i] = s0; dst[i+1] = s1;
            }
        }
        else
        {
            for( i = 0; i <= width - 2; i += 2 )
            {
                float s0 = kx[0]*src[0][i] + ky[0]*src[0][i+width];
                float s1 = kx[0]*src[0][i+1] + ky[0]*src[0][i+width+1];

                for( k = 1; k <= ksize2; k++ )
                {
                    const float* src1 = src[k] + i, *src2 = src[-k] + i;
                    float fx = kx[k], fy = ky[k];
                    s0 += fx*(src1[0] + src2[0]) + fy*(src1[width] + src2[width]);
                    s1 += fx*(src1[1] + src2[1]) + fy*(src1[width+1] + src2[width+1]);
                }

                s0 *= scale; s1 *= scale;
                dst[i] = s0; dst[i+1] = s1;
            }
        }

        for( ; i < width; i++ )
        {
            float s0 = kx[0]*src[0][i] + ky[0]*src[0][i+width];
            for( k = 1; k <= ksize2; k++ )
            {
                const float* src1 = src[k] + i, *src2 = src[-k] + i;
                s0 += kx[k]*(src1[0] + src2[0]) + ky[k]*(src1[width] + src2[width]);
            }
            dst[i] = s0*scale;
        }
    }
}


CV_IMPL void
cvLaplace( const void* srcarr, void* dstarr, int aperture_size )
{
    CvLaplaceFilter laplacian;
    void* buffer = 0;
    int local_alloc = 0;
    
    CV_FUNCNAME( "cvLaplace" );

    __BEGIN__;

    CvMat srcstub, *src = (CvMat*)srcarr;
    CvMat dststub, *dst = (CvMat*)dstarr;
    int src_type, dst_type;

    CV_CALL( src = cvGetMat( src, &srcstub ));
    CV_CALL( dst = cvGetMat( dst, &dststub ));

    src_type = CV_MAT_TYPE(src->type);
    dst_type = CV_MAT_TYPE(dst->type);

    if( aperture_size == 3 || aperture_size == 5 &&
        (src_type == CV_8UC1 && dst_type == CV_16SC1 ||
        src_type == CV_32FC1 && dst_type == CV_32FC1) &&
        (aperture_size == 3 || src_type < CV_32FC1) )
    {
        CvDerivGetBufSizeIPPFunc ipp_laplace_getbufsize_func = 0;
        CvDerivIPPFunc_8u ipp_laplace_func_8u = 0;
        CvDerivIPPFunc_32f ipp_laplace_func_32f = 0;

        if( src_type == CV_8U )
            ipp_laplace_func_8u = icvFilterLaplacianBorder_8u16s_C1R_p,
            ipp_laplace_getbufsize_func = icvFilterLaplacianGetBufSize_8u16s_C1R_p;
        else
            ipp_laplace_func_32f = icvFilterLaplacianBorder_32f_C1R_p,
            ipp_laplace_getbufsize_func = icvFilterLaplacianGetBufSize_32f_C1R_p;

        if( (ipp_laplace_func_8u || ipp_laplace_func_32f) && ipp_laplace_getbufsize_func )
        {
            int bufsize = 0, masksize = aperture_size == 3 ? 33 : 55;
            CvSize size = cvGetMatSize( src );
            uchar* src_ptr = src->data.ptr;
            uchar* dst_ptr = dst->data.ptr;
            int src_step = src->step ? src->step : CV_STUB_STEP;
            int dst_step = dst->step ? dst->step : CV_STUB_STEP;
            const int bordertype = 1; // replication border
            CvStatus status;

            status = ipp_laplace_getbufsize_func( size, masksize, &bufsize );

            if( status >= 0 )
            {
                if( bufsize <= CV_MAX_LOCAL_SIZE )
                {
                    buffer = cvStackAlloc( bufsize );
                    local_alloc = 1;
                }
                else
                    CV_CALL( buffer = cvAlloc( bufsize ));
            
                status =
                    ipp_laplace_func_8u ? ipp_laplace_func_8u( src_ptr, src_step, dst_ptr, dst_step,
                                                               size, masksize, bordertype, 0, buffer ) :
                    ipp_laplace_func_32f ? ipp_laplace_func_32f( src_ptr, src_step, dst_ptr, dst_step,
                                                                 size, masksize, bordertype, 0, buffer ) :
                        CV_NOTDEFINED_ERR;
            }

            if( status >= 0 )
                EXIT;
        }
    }

    CV_CALL( laplacian.init( src->cols, src_type, dst_type,
                             false, aperture_size ));
    CV_CALL( laplacian.process( src, dst ));

    __END__;

    if( buffer && !local_alloc )
        cvFree( &buffer );
}

/* End of file. */