1 | #include <iostream> |
---|
2 | #include <iomanip> |
---|
3 | #include <math.h> |
---|
4 | #include <duchamp.hh> |
---|
5 | #include <ATrous/atrous.hh> |
---|
6 | #include <Utils/utils.hh> |
---|
7 | #include <Utils/Statistics.hh> |
---|
8 | using Statistics::madfmToSigma; |
---|
9 | |
---|
10 | using std::endl; |
---|
11 | using std::setw; |
---|
12 | |
---|
13 | void atrous2DReconstruct(long &xdim, long &ydim, float *&input, float *&output, Param &par) |
---|
14 | { |
---|
15 | /** |
---|
16 | * atrous2DReconstruct(xdim, ydim, input, output, par) |
---|
17 | * |
---|
18 | * A routine that uses the a trous wavelet method to reconstruct a |
---|
19 | * 2-dimensional image. |
---|
20 | * The Param object "par" contains all necessary info about the filter and |
---|
21 | * reconstruction parameters, although a Filter object has to be declared |
---|
22 | * elsewhere previously. |
---|
23 | * The input array is in "input", of dimensions "xdim"x"ydim", and the |
---|
24 | * reconstructed array is in "output". |
---|
25 | */ |
---|
26 | |
---|
27 | extern Filter reconFilter; |
---|
28 | bool flagBlank=par.getFlagBlankPix(); |
---|
29 | float blankPixValue = par.getBlankPixVal(); |
---|
30 | long size = xdim * ydim; |
---|
31 | long mindim = xdim; |
---|
32 | if (ydim<mindim) mindim = ydim; |
---|
33 | int numScales = reconFilter.getNumScales(mindim); |
---|
34 | double *sigmaFactors = new double[numScales+1]; |
---|
35 | for(int i=0;i<=numScales;i++){ |
---|
36 | if(i<=reconFilter.maxFactor(2)) |
---|
37 | sigmaFactors[i] = reconFilter.sigmaFactor(2,i); |
---|
38 | else sigmaFactors[i] = sigmaFactors[i-1] / 2.; |
---|
39 | } |
---|
40 | |
---|
41 | float mean,sigma,originalSigma,originalMean,oldsigma,newsigma; |
---|
42 | bool *isGood = new bool[size]; |
---|
43 | for(int pos=0;pos<size;pos++) isGood[pos] = !par.isBlank(input[pos]); |
---|
44 | |
---|
45 | float *array = new float[size]; |
---|
46 | int goodSize=0; |
---|
47 | for(int i=0;i<size;i++) if(isGood[i]) array[goodSize++] = input[i]; |
---|
48 | findMedianStats(array,goodSize,originalMean,originalSigma); |
---|
49 | originalSigma = madfmToSigma(originalSigma); |
---|
50 | delete [] array; |
---|
51 | |
---|
52 | float *coeffs = new float[size]; |
---|
53 | float *wavelet = new float[size]; |
---|
54 | |
---|
55 | for(int pos=0;pos<size;pos++) output[pos]=0.; |
---|
56 | |
---|
57 | int filterwidth = reconFilter.width(); |
---|
58 | int filterHW = filterwidth/2; |
---|
59 | double *filter = new double[filterwidth*filterwidth]; |
---|
60 | for(int i=0;i<filterwidth;i++){ |
---|
61 | for(int j=0;j<filterwidth;j++){ |
---|
62 | filter[i*filterwidth+j] = reconFilter.coeff(i) * reconFilter.coeff(j); |
---|
63 | } |
---|
64 | } |
---|
65 | |
---|
66 | int *xLim1 = new int[ydim]; |
---|
67 | for(int i=0;i<ydim;i++) xLim1[i] = 0; |
---|
68 | int *yLim1 = new int[xdim]; |
---|
69 | for(int i=0;i<xdim;i++) yLim1[i] = 0; |
---|
70 | int *xLim2 = new int[ydim]; |
---|
71 | for(int i=0;i<ydim;i++) xLim2[i] = xdim-1; |
---|
72 | int *yLim2 = new int[xdim]; |
---|
73 | for(int i=0;i<xdim;i++) yLim2[i] = ydim-1; |
---|
74 | |
---|
75 | if(par.getFlagBlankPix()){ |
---|
76 | float avGapX = 0, avGapY = 0; |
---|
77 | for(int row=0;row<ydim;row++){ |
---|
78 | int ct1 = 0; |
---|
79 | int ct2 = xdim - 1; |
---|
80 | while((ct1<ct2)&&(par.isBlank(input[row*xdim+ct1]))) ct1++; |
---|
81 | while((ct2>ct1)&&(par.isBlank(input[row*xdim+ct2]))) ct2--; |
---|
82 | xLim1[row] = ct1; |
---|
83 | xLim2[row] = ct2; |
---|
84 | avGapX += ct2 - ct1; |
---|
85 | } |
---|
86 | avGapX /= float(ydim); |
---|
87 | |
---|
88 | for(int col=0;col<xdim;col++){ |
---|
89 | int ct1=0; |
---|
90 | int ct2=ydim-1; |
---|
91 | while((ct1<ct2)&&(par.isBlank(input[col+xdim*ct1]))) ct1++; |
---|
92 | while((ct2>ct1)&&(par.isBlank(input[col+xdim*ct2]))) ct2--; |
---|
93 | yLim1[col] = ct1; |
---|
94 | yLim2[col] = ct2; |
---|
95 | avGapY += ct2 - ct1; |
---|
96 | } |
---|
97 | avGapY /= float(xdim); |
---|
98 | |
---|
99 | mindim = int(avGapX); |
---|
100 | if(avGapY < avGapX) mindim = int(avGapY); |
---|
101 | numScales = reconFilter.getNumScales(mindim); |
---|
102 | } |
---|
103 | |
---|
104 | float threshold; |
---|
105 | int iteration=0; |
---|
106 | newsigma = 1.e9; |
---|
107 | for(int i=0;i<size;i++) output[i] = 0; |
---|
108 | do{ |
---|
109 | if(par.isVerbose()) { |
---|
110 | std::cout << "Iteration #"<<setw(2)<<++iteration<<":"; |
---|
111 | printBackSpace(13); |
---|
112 | } |
---|
113 | |
---|
114 | // first, get the value of oldsigma and set it to the previous |
---|
115 | // newsigma value |
---|
116 | oldsigma = newsigma; |
---|
117 | // we are transforming the residual array |
---|
118 | for(int i=0;i<size;i++) coeffs[i] = input[i] - output[i]; |
---|
119 | |
---|
120 | int spacing = 1; |
---|
121 | for(int scale = 1; scale<numScales; scale++){ |
---|
122 | |
---|
123 | if(par.isVerbose()){ |
---|
124 | std::cout << "Scale "; |
---|
125 | std::cout << setw(2)<<scale<<" / "<<setw(2)<<numScales; |
---|
126 | printBackSpace(13); |
---|
127 | std::cout <<std::flush; |
---|
128 | } |
---|
129 | |
---|
130 | for(int ypos = 0; ypos<ydim; ypos++){ |
---|
131 | for(int xpos = 0; xpos<xdim; xpos++){ |
---|
132 | // loops over each pixel in the image |
---|
133 | int pos = ypos*xdim + xpos; |
---|
134 | |
---|
135 | wavelet[pos] = coeffs[pos]; |
---|
136 | |
---|
137 | if(!isGood[pos]) wavelet[pos] = 0.; |
---|
138 | else{ |
---|
139 | |
---|
140 | int filterpos = -1; |
---|
141 | for(int yoffset=-filterHW; yoffset<=filterHW; yoffset++){ |
---|
142 | int y = ypos + spacing*yoffset; |
---|
143 | // Boundary conditions -- assume reflection at boundaries. |
---|
144 | // Use limits as calculated above |
---|
145 | // if(yLim1[xpos]!=yLim2[xpos]){ |
---|
146 | // // if these are equal we will get into an infinite loop here |
---|
147 | // while((y<yLim1[xpos])||(y>yLim2[xpos])){ |
---|
148 | // if(y<yLim1[xpos]) y = 2*yLim1[xpos] - y; |
---|
149 | // else if(y>yLim2[xpos]) y = 2*yLim2[xpos] - y; |
---|
150 | // } |
---|
151 | // } |
---|
152 | int oldrow = y * xdim; |
---|
153 | |
---|
154 | for(int xoffset=-filterHW; xoffset<=filterHW; xoffset++){ |
---|
155 | int x = xpos + spacing*xoffset; |
---|
156 | // Boundary conditions -- assume reflection at boundaries. |
---|
157 | // Use limits as calculated above |
---|
158 | // if(xLim1[ypos]!=xLim2[ypos]){ |
---|
159 | // // if these are equal we will get into an infinite loop here |
---|
160 | // while((x<xLim1[ypos])||(x>xLim2[ypos])){ |
---|
161 | // if(x<xLim1[ypos]) x = 2*xLim1[ypos] - x; |
---|
162 | // else if(x>xLim2[ypos]) x = 2*xLim2[ypos] - x; |
---|
163 | // } |
---|
164 | // } |
---|
165 | |
---|
166 | int oldpos = oldrow + x; |
---|
167 | |
---|
168 | float oldCoeff; |
---|
169 | if((y>=yLim1[xpos])&&(y<=yLim2[xpos])&& |
---|
170 | (x>=xLim1[ypos])&&(x<=xLim2[ypos]) ) |
---|
171 | oldCoeff = coeffs[oldpos]; |
---|
172 | else oldCoeff = 0.; |
---|
173 | |
---|
174 | filterpos++; |
---|
175 | |
---|
176 | if(isGood[pos]) wavelet[pos] -= filter[filterpos] * oldCoeff; |
---|
177 | // wavelet[pos] -= filter[filterpos] * coeffs[oldpos]; |
---|
178 | |
---|
179 | } //-> end of xoffset loop |
---|
180 | } //-> end of yoffset loop |
---|
181 | } //-> end of else{ ( from if(!isGood[pos]) ) |
---|
182 | |
---|
183 | } //-> end of xpos loop |
---|
184 | } //-> end of ypos loop |
---|
185 | |
---|
186 | // Need to do this after we've done *all* the convolving |
---|
187 | for(int pos=0;pos<size;pos++) coeffs[pos] = coeffs[pos] - wavelet[pos]; |
---|
188 | |
---|
189 | // Have found wavelet coeffs for this scale -- now threshold |
---|
190 | if(scale>=par.getMinScale()){ |
---|
191 | array = new float[size]; |
---|
192 | goodSize=0; |
---|
193 | for(int pos=0;pos<size;pos++){ |
---|
194 | if(isGood[pos]) array[goodSize++] = wavelet[pos]; |
---|
195 | } |
---|
196 | findMedianStats(array,goodSize,mean,sigma); |
---|
197 | delete [] array; |
---|
198 | |
---|
199 | threshold = mean + |
---|
200 | par.getAtrousCut() * originalSigma * sigmaFactors[scale]; |
---|
201 | for(int pos=0;pos<size;pos++){ |
---|
202 | if(!isGood[pos]) output[pos] = blankPixValue; |
---|
203 | // preserve the Blank pixel values in the output. |
---|
204 | else if( fabs(wavelet[pos]) > threshold ) |
---|
205 | output[pos] += wavelet[pos]; |
---|
206 | } |
---|
207 | } |
---|
208 | spacing *= 2; |
---|
209 | |
---|
210 | } // END OF LOOP OVER SCALES |
---|
211 | |
---|
212 | for(int pos=0;pos<size;pos++) if(isGood[pos]) output[pos] += coeffs[pos]; |
---|
213 | |
---|
214 | array = new float[size]; |
---|
215 | goodSize=0; |
---|
216 | for(int i=0;i<size;i++){ |
---|
217 | if(isGood[i]) array[goodSize++] = input[i] - output[i]; |
---|
218 | } |
---|
219 | findMedianStats(array,goodSize,mean,newsigma); |
---|
220 | newsigma = madfmToSigma(newsigma); |
---|
221 | delete [] array; |
---|
222 | |
---|
223 | if(par.isVerbose()) printBackSpace(15); |
---|
224 | |
---|
225 | } while( (iteration==1) || |
---|
226 | (fabs(oldsigma-newsigma)/newsigma > reconTolerance) ); |
---|
227 | |
---|
228 | if(par.isVerbose()) std::cout << "Completed "<<iteration<<" iterations. "; |
---|
229 | |
---|
230 | delete [] xLim1,xLim2,yLim1,yLim2; |
---|
231 | delete [] coeffs; |
---|
232 | delete [] wavelet; |
---|
233 | delete [] isGood; |
---|
234 | delete [] filter; |
---|
235 | delete [] sigmaFactors; |
---|
236 | } |
---|
237 | |
---|
238 | |
---|