1 | // ----------------------------------------------------------------------- |
---|
2 | // atrous_1d_reconstruct.cc: 1-dimensional wavelet reconstruction. |
---|
3 | // ----------------------------------------------------------------------- |
---|
4 | // Copyright (C) 2006, Matthew Whiting, ATNF |
---|
5 | // |
---|
6 | // This program is free software; you can redistribute it and/or modify it |
---|
7 | // under the terms of the GNU General Public License as published by the |
---|
8 | // Free Software Foundation; either version 2 of the License, or (at your |
---|
9 | // option) any later version. |
---|
10 | // |
---|
11 | // Duchamp is distributed in the hope that it will be useful, but WITHOUT |
---|
12 | // ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
---|
13 | // FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
---|
14 | // for more details. |
---|
15 | // |
---|
16 | // You should have received a copy of the GNU General Public License |
---|
17 | // along with Duchamp; if not, write to the Free Software Foundation, |
---|
18 | // Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307, USA |
---|
19 | // |
---|
20 | // Correspondence concerning Duchamp may be directed to: |
---|
21 | // Internet email: Matthew.Whiting [at] atnf.csiro.au |
---|
22 | // Postal address: Dr. Matthew Whiting |
---|
23 | // Australia Telescope National Facility, CSIRO |
---|
24 | // PO Box 76 |
---|
25 | // Epping NSW 1710 |
---|
26 | // AUSTRALIA |
---|
27 | // ----------------------------------------------------------------------- |
---|
28 | #include <iostream> |
---|
29 | #include <sstream> |
---|
30 | #include <iomanip> |
---|
31 | #include <math.h> |
---|
32 | #include <duchamp/duchamp.hh> |
---|
33 | #include <duchamp/param.hh> |
---|
34 | #include <duchamp/Utils/utils.hh> |
---|
35 | #include <duchamp/Utils/feedback.hh> |
---|
36 | #include <duchamp/ATrous/atrous.hh> |
---|
37 | #include <duchamp/ATrous/filter.hh> |
---|
38 | #include <duchamp/Utils/Statistics.hh> |
---|
39 | using Statistics::madfmToSigma; |
---|
40 | |
---|
41 | namespace duchamp |
---|
42 | { |
---|
43 | |
---|
44 | void atrous1DReconstruct(size_t &xdim, float *&input, float *&output, Param &par) |
---|
45 | { |
---|
46 | /// A routine that uses the a trous wavelet method to reconstruct a |
---|
47 | /// 1-dimensional spectrum. |
---|
48 | /// The Param object "par" contains all necessary info about the filter and |
---|
49 | /// reconstruction parameters. |
---|
50 | /// |
---|
51 | /// If all pixels are BLANK (and we are testing for BLANKs), the |
---|
52 | /// reconstruction will simply give BLANKs back, so we return the |
---|
53 | /// input array as the output array. |
---|
54 | /// |
---|
55 | /// \param xdim The length of the spectrum. |
---|
56 | /// \param input The input spectrum. |
---|
57 | /// \param output The returned reconstructed spectrum. This array needs to |
---|
58 | /// be declared beforehand. |
---|
59 | /// \param par The Param set. |
---|
60 | |
---|
61 | const float SNR_THRESH=par.getAtrousCut(); |
---|
62 | unsigned int MIN_SCALE=par.getMinScale(); |
---|
63 | unsigned int MAX_SCALE=par.getMaxScale(); |
---|
64 | static bool firstTime = true; // need this static in case we do two reconstructions - e.g. baseline subtraction |
---|
65 | |
---|
66 | unsigned int numScales = par.filter().getNumScales(xdim); |
---|
67 | if((MAX_SCALE>0)&&(MAX_SCALE<=numScales)) |
---|
68 | MAX_SCALE = std::min(MAX_SCALE,numScales); |
---|
69 | else{ |
---|
70 | if((firstTime)&&(MAX_SCALE!=0)){ |
---|
71 | firstTime=false; |
---|
72 | DUCHAMPWARN("Reading parameters","The requested value of the parameter scaleMax, \"" << par.getMaxScale() << "\" is outside the allowed range (1-"<< numScales <<") -- setting to " << numScales); |
---|
73 | } |
---|
74 | MAX_SCALE = numScales; |
---|
75 | par.setMaxScale(MAX_SCALE); |
---|
76 | } |
---|
77 | double *sigmaFactors = new double[numScales+1]; |
---|
78 | for(size_t i=0;i<=numScales;i++){ |
---|
79 | if(i<=par.filter().maxFactor(1)) |
---|
80 | sigmaFactors[i] = par.filter().sigmaFactor(1,i); |
---|
81 | else sigmaFactors[i] = sigmaFactors[i-1] / sqrt(2.); |
---|
82 | } |
---|
83 | |
---|
84 | float mean,originalSigma,oldsigma,newsigma; |
---|
85 | bool *isGood = new bool[xdim]; |
---|
86 | size_t goodSize=0; |
---|
87 | for(size_t pos=0;pos<xdim;pos++) { |
---|
88 | isGood[pos] = !par.isBlank(input[pos]); |
---|
89 | if(isGood[pos]) goodSize++; |
---|
90 | } |
---|
91 | |
---|
92 | if(goodSize == 0){ |
---|
93 | // There are no good pixels -- everything is BLANK for some reason. |
---|
94 | // Return the input array as the output. |
---|
95 | |
---|
96 | for(size_t pos=0;pos<xdim; pos++) output[pos] = input[pos]; |
---|
97 | |
---|
98 | } |
---|
99 | else{ |
---|
100 | // Otherwise, all is good, and we continue. |
---|
101 | |
---|
102 | |
---|
103 | float *coeffs = new float[xdim]; |
---|
104 | float *wavelet = new float[xdim]; |
---|
105 | // float *residual = new float[xdim]; |
---|
106 | |
---|
107 | for(size_t pos=0;pos<xdim;pos++) output[pos]=0.; |
---|
108 | |
---|
109 | int filterHW = par.filter().width()/2; |
---|
110 | double *filter = new double[par.filter().width()]; |
---|
111 | for(size_t i=0;i<par.filter().width();i++) filter[i] = par.filter().coeff(i); |
---|
112 | |
---|
113 | |
---|
114 | // No trimming done in 1D case. |
---|
115 | |
---|
116 | float threshold; |
---|
117 | int iteration=0; |
---|
118 | newsigma = 1.e9; |
---|
119 | do{ |
---|
120 | if(par.isVerbose()) { |
---|
121 | std::cout << "Iteration #"<<++iteration<<":"; |
---|
122 | printSpace(13); |
---|
123 | } |
---|
124 | // first, get the value of oldsigma and set it to the previous |
---|
125 | // newsigma value |
---|
126 | oldsigma = newsigma; |
---|
127 | // all other times round, we are transforming the residual array |
---|
128 | for(size_t i=0;i<xdim;i++) coeffs[i] = input[i] - output[i]; |
---|
129 | |
---|
130 | // findMedianStats(input,xdim,isGood,originalMean,originalSigma); |
---|
131 | // originalSigma = madfmToSigma(originalSigma); |
---|
132 | if(par.getFlagRobustStats()) |
---|
133 | originalSigma = madfmToSigma(findMADFM(input,isGood,xdim)); |
---|
134 | else |
---|
135 | originalSigma = findStddev<float>(input,isGood,xdim); |
---|
136 | |
---|
137 | int spacing = 1; |
---|
138 | for(unsigned int scale = 1; scale<=numScales; scale++){ |
---|
139 | |
---|
140 | if(par.isVerbose()) { |
---|
141 | std::cout << "Scale " << std::setw(2) << scale |
---|
142 | << " /" << std::setw(2) << numScales <<std::flush; |
---|
143 | } |
---|
144 | |
---|
145 | for(size_t xpos = 0; xpos<xdim; xpos++){ |
---|
146 | // loops over each pixel in the image |
---|
147 | |
---|
148 | wavelet[xpos] = coeffs[xpos]; |
---|
149 | |
---|
150 | if(!isGood[xpos] ) wavelet[xpos] = 0.; |
---|
151 | else{ |
---|
152 | |
---|
153 | for(int xoffset=-filterHW; xoffset<=filterHW; xoffset++){ |
---|
154 | long x = xpos + spacing*xoffset; |
---|
155 | |
---|
156 | while((x<0)||(x>=long(xdim))){ |
---|
157 | // boundary conditions are reflection. |
---|
158 | if(x<0) x = 0 - x; |
---|
159 | else if(x>=long(xdim)) x = 2*(xdim-1) - x; |
---|
160 | } |
---|
161 | |
---|
162 | size_t filterpos = (xoffset+filterHW); |
---|
163 | size_t oldpos = x; |
---|
164 | |
---|
165 | if(isGood[oldpos]) |
---|
166 | wavelet[xpos] -= filter[filterpos]*coeffs[oldpos]; |
---|
167 | |
---|
168 | } //-> end of xoffset loop |
---|
169 | } //-> end of else{ ( from if(!isGood[xpos]) ) |
---|
170 | |
---|
171 | } //-> end of xpos loop |
---|
172 | |
---|
173 | // Need to do this after we've done *all* the convolving |
---|
174 | for(size_t pos=0;pos<xdim;pos++) coeffs[pos] = coeffs[pos] - wavelet[pos]; |
---|
175 | |
---|
176 | // Have found wavelet coeffs for this scale -- now threshold |
---|
177 | if(scale>=MIN_SCALE && scale <=MAX_SCALE){ |
---|
178 | // findMedianStats(wavelet,xdim,isGood,mean,sigma); |
---|
179 | if(par.getFlagRobustStats()) |
---|
180 | mean = findMedian<float>(wavelet,isGood,xdim); |
---|
181 | else |
---|
182 | mean = findMean<float>(wavelet,isGood,xdim); |
---|
183 | |
---|
184 | threshold = mean+SNR_THRESH*originalSigma*sigmaFactors[scale]; |
---|
185 | for(size_t pos=0;pos<xdim;pos++){ |
---|
186 | // preserve the Blank pixel values in the output. |
---|
187 | if(!isGood[pos]) output[pos] = input[pos]; |
---|
188 | else if( fabs(wavelet[pos]) > threshold ) |
---|
189 | output[pos] += wavelet[pos]; |
---|
190 | } |
---|
191 | } |
---|
192 | |
---|
193 | spacing *= 2; |
---|
194 | |
---|
195 | } //-> end of scale loop |
---|
196 | |
---|
197 | // Only add the final smoothed array if we are doing *all* the scales. |
---|
198 | if(numScales == par.filter().getNumScales(xdim)) |
---|
199 | for(size_t pos=0;pos<xdim;pos++) |
---|
200 | if(isGood[pos]) output[pos] += coeffs[pos]; |
---|
201 | |
---|
202 | // for(size_t pos=0;pos<xdim;pos++) residual[pos]=input[pos]-output[pos]; |
---|
203 | // findMedianStats(residual,xdim,isGood,mean,newsigma); |
---|
204 | // newsigma = madfmToSigma(newsigma); |
---|
205 | if(par.getFlagRobustStats()) |
---|
206 | newsigma = madfmToSigma(findMADFMDiff(input,output,isGood,xdim)); |
---|
207 | else |
---|
208 | newsigma = findStddevDiff<float>(input,output,isGood,xdim); |
---|
209 | |
---|
210 | if(par.isVerbose()) printBackSpace(26); |
---|
211 | |
---|
212 | } while( (iteration==1) || |
---|
213 | (fabs(oldsigma-newsigma)/newsigma > par.getReconConvergence()) ); |
---|
214 | |
---|
215 | if(par.isVerbose()) std::cout << "Completed "<<iteration<<" iterations. "; |
---|
216 | |
---|
217 | delete [] filter; |
---|
218 | // delete [] residual; |
---|
219 | delete [] wavelet; |
---|
220 | delete [] coeffs; |
---|
221 | |
---|
222 | } |
---|
223 | |
---|
224 | delete [] isGood; |
---|
225 | delete [] sigmaFactors; |
---|
226 | } |
---|
227 | |
---|
228 | } |
---|