source: trunk/src/ATrous/ReconSearch.cc @ 146

#include <fstream>
#include <iostream>
#include <iomanip>
#include <vector>
#include <duchamp.hh>
#include <Cubes/cubes.hh>
#include <Detection/detection.hh>
#include <ATrous/atrous.hh>
#include <Utils/utils.hh>

using std::setw;

/////////////////////////////////////////////////////////////////////////////
void Cube::ReconSearch()
{
  /**
   * Cube::ReconSearch()
   *   A front-end to the various ReconSearch functions, the choice of 
   *   which is determined by the use of the reconDim parameter.
   */
  int dimRecon = this->par.getReconDim();
  
  // Test whether we have eg. an image, but have requested a 3-d 
  //  reconstruction.
  // If dimension of data array is less than dimRecon, change dimRecon 
  //  to the dimension of the array.
  int numGoodDim = 0;
  for(int i=0;i<this->numDim;i++) if(this->axisDim[i]>1) numGoodDim++;
  if(numGoodDim<dimRecon) dimRecon = numGoodDim;
  this->par.setReconDim(dimRecon);

  switch(dimRecon)
    {
    case 1: this->ReconSearch1D(); break;
    case 2: this->ReconSearch2D(); break;
    case 3: this->ReconSearch3D(); break;
    default:
      if(dimRecon<=0){
        std::stringstream errmsg;
        errmsg << "reconDim (" << dimRecon
               << ") is less than 1. Performing 1-D reconstruction.\n";
        duchampWarning("ReconSearch", errmsg.str());
        this->par.setReconDim(1);
        this->ReconSearch1D();
      }
      else if(dimRecon>3){ 
        //this probably won't happen with new code above, but just in case...
        std::stringstream errmsg;
        errmsg << "reconDim (" << dimRecon
               << ") is more than 3. Performing 3-D reconstruction.\n";
        duchampWarning("ReconSearch", errmsg.str());
        this->par.setReconDim(3);
        this->ReconSearch3D();
      }
      break;
    }
  
}


/////////////////////////////////////////////////////////////////////////////
void Cube::ReconSearch1D()
{
  /**
   * Cube::ReconSearch1D()
   *   This reconstructs a cube by performing a 1D a trous reconstruction
   *   in the spectrum of each spatial pixel.
   *   It then searches the cube using searchReconArray (below).
   *
   *   The resulting object list is stored in the Cube.
   */
  long xySize = this->axisDim[0] * this->axisDim[1];
  long zdim = this->axisDim[2];

  // Reconstruct the cube by 1d atrous transform in each spatial pixel
  if(!this->reconExists){
    std::cout<<"  Reconstructing... ";
    if(par.isVerbose()) std::cout << "|                    |" << std::flush;
    for(int npix=0; npix<xySize; npix++){
      if( par.isVerbose() && ((100*(npix+1)/xySize)%5 == 0) ){
        std::cout << "\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b|";
        for(int i=0;i<(100*(npix+1)/xySize)/5;i++) std::cout << "#";
        for(int i=(100*(npix+1)/xySize)/5;i<20;i++) std::cout << " ";
        std::cout << "|" << std::flush;
      }
      float *spec = new float[zdim];
      float *newSpec = new float[zdim];
      for(int z=0;z<zdim;z++) spec[z] = this->array[z*xySize + npix];
      bool verboseFlag = this->par.isVerbose();
      this->par.setVerbosity(false);
      atrous1DReconstruct(this->axisDim[2],spec,newSpec,this->par);
      this->par.setVerbosity(verboseFlag);
      for(int z=0;z<zdim;z++) this->recon[z*xySize+npix] = newSpec[z];
      delete spec;
      delete newSpec;
    }
    this->reconExists = true;
    std::cout<<"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b"
             <<"  All Done.                      \n"
             <<"  Searching... "<<std::flush;
  }
    
  this->objectList = searchReconArray(this->axisDim,this->array,
                                      this->recon,this->par);

  this->updateDetectMap();
  if(this->par.getFlagLog()) this->logDetectionList();

}

/////////////////////////////////////////////////////////////////////////////
void Cube::ReconSearch2D()
{
  /**
   * Cube::ReconSearch2D()
   *   This reconstructs a cube by performing a 2D a trous reconstruction
   *   in each spatial image (ie. each channel map) of the cube.
   *   It then searches the cube using searchReconArray (below).
   *
   *   The resulting object list is stored in the Cube.
   */
  long xySize = this->axisDim[0] * this->axisDim[1];

  if(!this->reconExists){
    // RECONSTRUCT THE CUBE BY 2D ATROUS TRANSFORM IN EACH CHANNEL  
    std::cout<<"  Reconstructing... ";
    if(par.isVerbose()) std::cout << "|                    |" << std::flush;
    for(int z=0;z<this->axisDim[2];z++){
      if( par.isVerbose() && ((100*(z+1)/this->axisDim[2])%5 == 0) ){
        std::cout << "\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b|";
        for(int i=0;i<(100*(z+1)/this->axisDim[2])/5;i++) std::cout << "#";
        for(int i=(100*(z+1)/this->axisDim[2])/5;i<20;i++) std::cout << " ";
        std::cout << "|" << std::flush;
      }
      if(!this->par.isInMW(z)){
        float *im = new float[xySize];
        float *newIm = new float[xySize];
        for(int npix=0; npix<xySize; npix++) 
          im[npix] = this->array[z*xySize+npix];
        bool verboseFlag = this->par.isVerbose();
        this->par.setVerbosity(false);
        atrous2DReconstruct(this->axisDim[0],this->axisDim[1],
                            im,newIm,this->par);
        this->par.setVerbosity(verboseFlag);
        for(int npix=0; npix<xySize; npix++) 
          this->recon[z*xySize+npix] = newIm[npix];
        delete im;
        delete newIm;
      }
      else {
        for(int i=z*xySize; i<(z+1)*xySize; i++) 
          this->recon[i] = this->array[i];
      }
    }
    this->reconExists = true;
    std::cout<<"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b"
             <<"  All Done.                      \n"
             <<"Searching... "<<std::flush;

  }

  this->objectList = searchReconArray(this->axisDim,this->array,
                                      this->recon,this->par);
  
  this->updateDetectMap();
  if(this->par.getFlagLog()) this->logDetectionList();

}

/////////////////////////////////////////////////////////////////////////////
void Cube::ReconSearch3D()
{
  /**
   * Cube::ReconSearch3D()
   *   This performs a full 3D a trous reconstruction of the cube
   *   It then searches the cube using searchReconArray (below).
   *
   *   The resulting object list is stored in the Cube.
   */
  if(this->axisDim[2]==1) this->ReconSearch2D();
  else {

    if(!this->reconExists){
      std::cout<<"  Reconstructing... "<<std::flush;
      atrous3DReconstruct(this->axisDim[0],this->axisDim[1],this->axisDim[2],
                          this->array,this->recon,this->par);
      this->reconExists = true;
      std::cout << "All Done.                      \n  Searching... "
                << std::flush;
    }

    this->objectList = searchReconArray(this->axisDim,this->array,
                                        this->recon,this->par);

    this->updateDetectMap();
    if(this->par.getFlagLog()) this->logDetectionList();

  }

}


/////////////////////////////////////////////////////////////////////////////
vector <Detection> searchReconArray(long *dim, float *originalArray, 
                                    float *reconArray, Param &par)
{
  /**
   * searchReconArray(long *dim, float *originalArray, 
   *                  float *reconArray, Param &par)
   *   This searches for objects in a cube that has been reconstructed.
   *
   *   Inputs:   - dimension array
   *             - original, un-reconstructed image array
   *             - reconstructed image array
   *             - parameters
   *
   *   Searches first in each spatial pixel (1D search), 
   *   then in each channel image (2D search).
   *
   *   Returns: vector of Detections resulting from the search.
   */
  vector <Detection> outputList;
  long zdim = dim[2];
  long xySize = dim[0] * dim[1];
  long fullSize = zdim * xySize;
  int num=0, goodSize;

  float blankPixValue = par.getBlankPixVal();
  bool *isGood = new bool[fullSize];
  for(int pos=0;pos<fullSize;pos++){
    isGood[pos] = !par.isBlank(originalArray[pos]) && !par.isInMW(pos/xySize);
  }
  bool *doChannel = new bool[xySize];
  goodSize=0;
  for(int npix=0; npix<xySize; npix++){
    for(int z=0;z<zdim;z++) if(isGood[z*xySize+npix]) goodSize++;
    if(goodSize==0) doChannel[npix] = false;
    else doChannel[npix] = true;
  }
 
  float dud;

  // First search --  in each spectrum.
  if(zdim > 1){
    if(par.isVerbose()) std::cout << "1D: |                    |" 
                                  << std::flush;

    for(int npix=0; npix<xySize; npix++){

      if( par.isVerbose() && ((100*(npix+1)/xySize)%5 == 0) ){
        std::cout << "\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b|";
        for(int i=0;i<(100*(npix+1)/xySize)/5;i++) std::cout << "#";
        for(int i=(100*(npix+1)/xySize)/5;i<20;i++) std::cout << " ";
        std::cout << "|" << std::flush;
      }

      if(doChannel[npix]){
        
        // First, get stats
        float *spec = new float[zdim];
        float specMedian,specSigma;
        goodSize=0;
        for(int z=0;z<zdim;z++) {
          if(isGood[z*xySize+npix]) 
            spec[goodSize++] = originalArray[z*xySize+npix];
        }
        specMedian = findMedian(spec,goodSize);
        goodSize=0;
        for(int z=0;z<zdim;z++) {
          if(isGood[z*xySize+npix]) 
            spec[goodSize++] = originalArray[z*xySize+npix] -
              reconArray[z*xySize+npix];
        }
        specSigma = findMADFM(spec,goodSize) / correctionFactor;
        delete [] spec;
        
        // Next, do source finding.
        long *specdim = new long[2];
        specdim[0] = zdim; specdim[1]=1;
        Image *spectrum = new Image(specdim);
        delete [] specdim;
        spectrum->saveParam(par);
        spectrum->pars().setBeamSize(2.); 
        // beam size: for spectrum, only neighbouring channels correlated
        spectrum->extractSpectrum(reconArray,dim,npix);
        spectrum->removeMW(); // only works if flagMW is true
        spectrum->setStats(specMedian,specSigma,par.getCut());
        if(par.getFlagFDR()) spectrum->setupFDR();
        spectrum->setMinSize(par.getMinChannels());
        spectrum->spectrumDetect();
        num += spectrum->getNumObj();
        for(int obj=0;obj<spectrum->getNumObj();obj++){
          Detection *object = new Detection;
          *object = spectrum->getObject(obj);
          for(int pix=0;pix<object->getSize();pix++) {
            // Fix up coordinates of each pixel to match original array
            object->setZ(pix, object->getX(pix));
            object->setX(pix, npix%dim[0]);
            object->setY(pix, npix/dim[0]);
            object->setF(pix, originalArray[object->getX(pix)+
                                            object->getY(pix)*dim[0]+
                                            object->getZ(pix)*xySize]); 
            // NB: set F to the original value, not the recon value.
          }
          object->addOffsets(par);
          object->calcParams();
          mergeIntoList(*object,outputList,par);
          delete object;
        }
        delete spectrum;
      }
    }

    num = outputList.size();
    if(par.isVerbose()) 
      std::cout <<"\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\bFound " 
                << num <<"; " << std::flush;

  }

  // Second search --  in each channel
  if(par.isVerbose()) std::cout << "2D: |                    |" 
                                << std::flush;

  num = 0;

  for(int z=0; z<zdim; z++){  // loop over all channels

    if( par.isVerbose() && ((100*(z+1)/zdim)%5 == 0) ){
      std::cout << "\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b|";
      for(int i=0;i<(100*(z+1)/zdim)/5;i++)  std::cout << "#";
      for(int i=(100*(z+1)/zdim)/5;i<20;i++) std::cout << " ";
      std::cout << "|" << std::flush;
    }

    if(!par.isInMW(z)){ 
      // purpose of this is to ignore the Milky Way channels 
      //  if we are flagging them

      //  First, get stats
      float *image = new float[xySize];
      float imageMedian,imageSigma;
      goodSize=0;
      for(int npix=0; npix<xySize; npix++) {
        if(isGood[z*xySize + npix]) 
          image[goodSize++] = originalArray[z*xySize + npix];
      }
      imageMedian = findMedian(image,goodSize);
      goodSize=0;
      for(int npix=0; npix<xySize; npix++) 
        if(isGood[z*xySize+npix]) 
          image[goodSize++]=originalArray[z*xySize+npix]-
            reconArray[z*xySize+npix];
      imageSigma = findMADFM(image,goodSize) / correctionFactor;
      delete [] image;

      // Next, do source finding.
      long *imdim = new long[2];
      imdim[0] = dim[0]; imdim[1] = dim[1];
      Image *channelImage = new Image(imdim);
      channelImage->saveParam(par);
      delete [] imdim;
      channelImage->extractImage(reconArray,dim,z);
      channelImage->setStats(imageMedian,imageSigma,par.getCut());
      if(par.getFlagFDR()) channelImage->setupFDR();
      channelImage->setMinSize(par.getMinPix());
      channelImage->lutz_detect();
      num += channelImage->getNumObj();
      for(int obj=0;obj<channelImage->getNumObj();obj++){
        Detection *object = new Detection;
        *object = channelImage->getObject(obj);
        // Fix up z coordinates of each pixel to match original array 
        //   (x & y are fine)
        for(int pix=0;pix<object->getSize();pix++){
          object->setZ(pix, z);
          object->setF(pix, originalArray[object->getX(pix)+
                                          object->getY(pix)*dim[0]+
                                          z*xySize]); 
          // NB: set F to the original value, not the recon value.
        }
        object->addOffsets(par);
        object->calcParams();
        mergeIntoList(*object,outputList,par);
        delete object;
      }
      delete channelImage;
   }

  }


  std::cout << "\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\b\bFound " 
            << num 
            << ".                                           "  
            << std::endl << std::flush;

  delete [] isGood;
  delete [] doChannel;

  return outputList;
}