source: trunk/src/ATrous/atrous_3d_reconstruct.cc

// -----------------------------------------------------------------------
// atrous_3d_reconstruct.cc: 3-dimensional wavelet reconstruction.
// -----------------------------------------------------------------------
// Copyright (C) 2006, Matthew Whiting, ATNF
//
// This program is free software; you can redistribute it and/or modify it
// under the terms of the GNU General Public License as published by the
// Free Software Foundation; either version 2 of the License, or (at your
// option) any later version.
//
// Duchamp is distributed in the hope that it will be useful, but WITHOUT
// ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
// FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
// for more details.
//
// You should have received a copy of the GNU General Public License
// along with Duchamp; if not, write to the Free Software Foundation,
// Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307, USA
//
// Correspondence concerning Duchamp may be directed to:
//    Internet email: Matthew.Whiting [at] atnf.csiro.au
//    Postal address: Dr. Matthew Whiting
//                    Australia Telescope National Facility, CSIRO
//                    PO Box 76
//                    Epping NSW 1710
//                    AUSTRALIA
// -----------------------------------------------------------------------
#include <iostream>
#include <iomanip>
#include <math.h>
#include <duchamp/duchamp.hh>
#include <duchamp/param.hh>
#include <duchamp/ATrous/atrous.hh>
#include <duchamp/ATrous/filter.hh>
#include <duchamp/Utils/utils.hh>
#include <duchamp/Utils/feedback.hh>
#include <duchamp/Utils/Statistics.hh>
using Statistics::madfmToSigma;

using std::endl;
using std::setw;

namespace duchamp
{

  void atrous3DReconstruct(size_t &xdim, size_t &ydim, size_t &zdim, float *&input, 
                           float *&output, Param &par)
  {
    ///  A routine that uses the a trous wavelet method to reconstruct a 
    ///   3-dimensional image cube.
    ///  The Param object "par" contains all necessary info about the filter and 
    ///   reconstruction parameters.
    /// 
    ///  If there are no non-BLANK pixels (and we are testing for
    ///  BLANKs), the reconstruction cannot be done, so we return the
    ///  input array as the output array and give a warning message.
    /// 
    ///  \param xdim The length of the x-axis.
    ///  \param ydim The length of the y-axis.
    ///  \param zdim The length of the z-axis.
    ///  \param input The input spectrum.
    ///  \param output The returned reconstructed spectrum. This array needs to be declared beforehand.
    ///  \param par The Param set.

    const float SNR_THRESH=par.getAtrousCut();
    unsigned int MIN_SCALE=par.getMinScale();
    unsigned int MAX_SCALE=par.getMaxScale();

    size_t size = xdim * ydim * zdim;
    size_t spatialSize = xdim * ydim;
    size_t mindim = xdim;
    if (ydim<mindim) mindim = ydim;
    if (zdim<mindim) mindim = zdim;

    unsigned int numScales = par.filter().getNumScales(mindim);
    if((MAX_SCALE>0)&&(MAX_SCALE<=numScales))
      MAX_SCALE = std::min(MAX_SCALE,numScales);
    else{
      if(MAX_SCALE!=0)
        DUCHAMPWARN("Reading parameters","The requested value of the parameter scaleMax, \"" << par.getMaxScale() << "\" is outside the allowed range (1-"<< numScales <<") -- setting to " << numScales);
      MAX_SCALE = numScales;
      par.setMaxScale(MAX_SCALE);
    }

    double *sigmaFactors = new double[numScales+1];
    for(size_t i=0;i<=numScales;i++){
      if(i<=size_t(par.filter().maxFactor(3)) )
        sigmaFactors[i] = par.filter().sigmaFactor(3,i);
      else sigmaFactors[i] = sigmaFactors[i-1] / sqrt(8.);
    }

    float mean,originalSigma,oldsigma,newsigma;
    std::vector<bool> isGood(size);
    size_t goodSize=0;
    for(size_t pos=0;pos<size;pos++){
      isGood[pos] = !par.isBlank(input[pos]);
      if(isGood[pos]) goodSize++;
    }

    if(goodSize == 0){
      // There are no good pixels -- everything is BLANK for some reason.
      // Return the input array as the output, and give a warning message.

      for(size_t pos=0;pos<xdim; pos++) output[pos] = input[pos];

      DUCHAMPWARN("3D Reconstruction", "There are no good pixels to be reconstructed -- all are BLANK. Returning input array.\n");
    }
    else{
      // Otherwise, all is good, and we continue.


      // findMedianStats(input,goodSize,isGood,originalMean,originalSigma);
      if(par.getFlagRobustStats())
        originalSigma = madfmToSigma(findMADFM(input,isGood,size));
      else
        originalSigma = findStddev<float>(input,isGood,size);

      float *coeffs = new float[size];
      float *wavelet = new float[size];
      // float *residual = new float[size];

      for(size_t pos=0;pos<size;pos++) output[pos]=0.;

      // Define the 3-D (separable) filter, using info from par.filter()
      size_t filterwidth = par.filter().width();
      int filterHW = filterwidth/2;
      size_t fsize = filterwidth*filterwidth*filterwidth;
      double *filter = new double[fsize];
      for(size_t i=0;i<filterwidth;i++){
        for(size_t j=0;j<filterwidth;j++){
          for(size_t k=0;k<filterwidth;k++){
            filter[i +j*filterwidth + k*filterwidth*filterwidth] = 
              par.filter().coeff(i) * par.filter().coeff(j) * par.filter().coeff(k);
          }
        }
      }

      // // Locating the borders of the image -- ignoring BLANK pixels
      // //  Only do this if flagBlankPix is true. 
      // //  Otherwise use the full range of x and y.
      // //  No trimming is done in the z-direction at this point.
      // long *xLim1 = new long[ydim];
      // for(size_t i=0;i<ydim;i++) xLim1[i] = 0;
      // long *xLim2 = new long[ydim];
      // for(size_t i=0;i<ydim;i++) xLim2[i] = xdim-1;
      // long *yLim1 = new long[xdim];
      // for(size_t i=0;i<xdim;i++) yLim1[i] = 0;
      // long *yLim2 = new long[xdim];
      // for(size_t i=0;i<xdim;i++) yLim2[i] = ydim-1;

      // if(par.getFlagBlankPix()){
      //        float avGapX = 0, avGapY = 0;
      //        for(size_t row=0;row<ydim;row++){
      //          size_t ct1 = 0;
      //          size_t ct2 = xdim - 1;
      //          while((ct1<ct2)&&(par.isBlank(input[row*xdim+ct1]))) ct1++;
      //          while((ct2>ct1)&&(par.isBlank(input[row*xdim+ct2]))) ct2--;
      //          xLim1[row] = ct1;
      //          xLim2[row] = ct2;
      //          avGapX += ct2 - ct1 + 1;
      //        }
      //        avGapX /= float(ydim);

      //        for(size_t col=0;col<xdim;col++){
      //          size_t ct1=0;
      //          size_t ct2=ydim-1;
      //          while((ct1<ct2)&&(par.isBlank(input[col+xdim*ct1]))) ct1++;
      //          while((ct2>ct1)&&(par.isBlank(input[col+xdim*ct2]))) ct2--;
      //          yLim1[col] = ct1;
      //          yLim2[col] = ct2;
      //          avGapY += ct2 - ct1 + 1;
      //        }
      //        avGapY /= float(xdim);
 
      //        // if(avGapX < mindim) mindim = int(avGapX);
      //        // if(avGapY < mindim) mindim = int(avGapY);
      //        // numScales = par.filter().getNumScales(mindim);
      // }

      float threshold;
      int iteration=0;
      newsigma = 1.e9;
      for(size_t i=0;i<size;i++) output[i] = 0;
      do{
        if(par.isVerbose()) std::cout << "Iteration #"<<setw(2)<<++iteration<<": ";
        // first, get the value of oldsigma, set it to the previous newsigma value
        oldsigma = newsigma;
        // we are transforming the residual array (input array first time around)
        for(size_t i=0;i<size;i++)  coeffs[i] = input[i] - output[i];

        int spacing = 1;
        for(unsigned int scale = 1; scale<=numScales; scale++){

          if(par.isVerbose()){
            std::cout << "Scale ";
            std::cout << setw(2)<<scale<<" / "<<setw(2)<<numScales;
            printBackSpace(std::cout,13);
            std::cout << std::flush;
          }

          size_t pos = 0;
          for(unsigned long zpos = 0; zpos<zdim; zpos++){
            for(unsigned long ypos = 0; ypos<ydim; ypos++){
              for(unsigned long xpos = 0; xpos<xdim; xpos++){
                // loops over each pixel in the image

                wavelet[pos] = coeffs[pos];
            
                if(!isGood[pos] )  wavelet[pos] = 0.;
                else{

                  unsigned int filterpos = 0;
                  for(int zoffset=-filterHW; zoffset<=filterHW; zoffset++){
                    long z = zpos + spacing*zoffset;
                    while((z<0)||(z>=long(zdim))){
                      if(z<0) z = -z;                 // boundary conditions are 
                      if(z>=long(zdim)) z = 2*(zdim-1) - z; //    reflection.
                    }
                    size_t oldchan = z * spatialSize;
                
                    for(int yoffset=-filterHW; yoffset<=filterHW; yoffset++){
                      long y = long(ypos) + spacing*yoffset;
                      while((y<0)||(y>=long(ydim))){
                        // boundary conditions are reflection. 
                        if(y<0) y = 0 - y;
                        else if(y>=long(ydim)) y = 2*(ydim-1) - y;
                      }

                      // // Boundary conditions -- assume reflection at boundaries.
                      // // Use limits as calculated above
                      // if(yLim1[xpos]!=yLim2[xpos]){ 
                      //        // if these are equal we will get into an infinite loop
                      //        while((y<yLim1[xpos])||(y>yLim2[xpos])){
                      //        // std::cerr << y << " " <<spacing << " " << yoffset << " " << ypos << " " << xpos << " " << yLim1[xpos] << " " << yLim2[xpos] << "\n";
                      //          if(y<yLim1[xpos]) y = 2*yLim1[xpos] - y;      
                      //          else if(y>yLim2[xpos]) y = 2*yLim2[xpos] - y;      
                      //        }
                      // }
                      size_t oldrow = y * xdim;
          
                      for(int xoffset=-filterHW; xoffset<=filterHW; xoffset++){
                        long x = long(xpos) + spacing*xoffset;
                        while((x<0)||(x>=long(xdim))){
                          // boundary conditions are reflection. 
                          if(x<0) x = 0 - x;
                          else if(x>=long(xdim)) x = 2*(xdim-1) - x;
                        }

                        // // Boundary conditions -- assume reflection at boundaries.
                        // // Use limits as calculated above
                        // if(xLim1[ypos]!=xLim2[ypos]){
                        //   // if these are equal we will get into an infinite loop
                        //   while((x<xLim1[ypos])||(x>xLim2[ypos])){
                        //     if(x<xLim1[ypos]) x = 2*xLim1[ypos] - x;      
                        //     else if(x>xLim2[ypos]) x = 2*xLim2[ypos] - x;      
                        //   }
                        // }

                        size_t oldpos = oldchan + oldrow + x;

                        if(isGood[oldpos]) 
                          wavelet[pos] -= filter[filterpos]*coeffs[oldpos];
                      
                        filterpos++;
                   
                      } //-> end of xoffset loop
                    } //-> end of yoffset loop
                  } //-> end of zoffset loop
                } //-> end of else{ ( from if(!isGood[pos])  )
            
                pos++;
              } //-> end of xpos loop
            } //-> end of ypos loop
          } //-> end of zpos loop

          // Need to do this after we've done *all* the convolving
          for(size_t pos=0;pos<size;pos++) coeffs[pos] = coeffs[pos] - wavelet[pos];

          // Have found wavelet coeffs for this scale -- now threshold
          if(scale>=MIN_SCALE && scale <=MAX_SCALE){
            if(par.getFlagRobustStats())
              // findMedianStats(wavelet,size,isGood,mean,sigma);
              mean = findMedian<float>(wavelet,isGood,size);
            else
              //findNormalStats(wavelet,size,isGood,mean,sigma);
              mean = findMean<float>(wavelet,isGood,size);
              
            threshold = mean + SNR_THRESH*originalSigma*sigmaFactors[scale];
            for(size_t pos=0;pos<size;pos++){
              if(!isGood[pos]){
                output[pos] = input[pos]; 
                // this preserves the Blank pixel values in the output.
              }
              else if( fabs(wavelet[pos]) > threshold ){
                output[pos] += wavelet[pos];
                // only add to the output if the wavelet coefficient is significant
              }
            }
          }
 
          spacing *= 2;  // double the scale of the filter.

        } //-> end of scale loop 

        for(size_t pos=0;pos<size;pos++) if(isGood[pos]) output[pos] += coeffs[pos];

        // for(size_t i=0;i<size;i++) residual[i] = input[i] - output[i];
        // findMedianStats(residual,goodSize,isGood,mean,newsigma);
        // findMedianStatsDiff(input,output,goodSize,isGood,mean,newsigma);
        // newsigma = madfmToSigma(newsigma);
        if(par.getFlagRobustStats())
          newsigma = madfmToSigma(findMADFMDiff(input,output,isGood,size));
        else
          newsigma = findStddevDiff<float>(input,output,isGood,size);

        if(par.isVerbose()) printBackSpace(std::cout,15);

      } while( (iteration==1) || 
               (fabs(oldsigma-newsigma)/newsigma > par.getReconConvergence()) );

      if(par.isVerbose()) std::cout << "Completed "<<iteration<<" iterations. ";

      // delete [] xLim1;
      // delete [] xLim2;
      // delete [] yLim1;
      // delete [] yLim2;
      delete [] filter;
      // delete [] residual;
      delete [] coeffs;
      delete [] wavelet;

    }

    delete [] sigmaFactors;
  }

}