Skip to content

I
imagej-elphel
Commit 1671728b authored by Andrey Filippov's avatar Andrey Filippov
Browse files
Options

Equalizing intensities

parent c4df622d
No related merge requests found
Expand all Show whitespace changes
Inline Side-by-side
Showing with 916 additions and 583 deletions
src/main/java/com/elphel/imagej/common/PolynomialApproximation.java
View file @ 1671728b
package com.elphel.imagej.common;
import java.util.concurrent.atomic.AtomicInteger;
import com.elphel.imagej.tileprocessor.ImageDtt;
import com.elphel.imagej.tileprocessor.QuadCLT;
import Jama.LUDecomposition;
import Jama.Matrix;
public class PolynomialApproximation {
public int debugLevel=1;
// TODO Move other methods here
// TODO Move other methods here
public PolynomialApproximation(){}
public PolynomialApproximation(int debugLevel){
this.debugLevel=debugLevel;
......@@ -53,7 +58,7 @@ public class PolynomialApproximation {
System.out.println("polynomialApproximation1d() B:");
B.print(10, 5);
}
// while (!(new LUDecomposition(M)).isNonsingular() && (N1>0)){
// while (!(new LUDecomposition(M)).isNonsingular() && (N1>0)){
while (!(new LUDecomposition(M)).isNonsingular() && (N1>=0)){ // make N=0 legal ?
aM=new double [N1][N1];
aB=new double [N1][1];
......@@ -81,7 +86,7 @@ public class PolynomialApproximation {
for (int i=0;i<=N;i++) result[i]=(i<=N1)?aR[i][0]:0.0;
return result;
}
/**
/**
* Linear approximates each of 3 functions of 3 variables and finds where they are all zero
* @param data: for each sample (1-st index):
* 0 - {x,y,z}
......@@ -312,7 +317,7 @@ public class PolynomialApproximation {
1.0E-15, // threshold ratio of matrix determinant to norm for quadratic approximation (det too low - fail)
this.debugLevel);
}
/*
/*
public double [][] quadraticApproximation( // no use
double [][][] data,
boolean forceLinear, // use linear approximation
......@@ -328,7 +333,7 @@ public class PolynomialApproximation {
1.0E-15, // threshold ratio of matrix determinant to norm for quadratic approximation (det too low - fail)
this.debugLevel);
}
*/
*/
public double [][] quadraticApproximation(
double [][][] data,
boolean forceLinear, // use linear approximation
......@@ -528,7 +533,7 @@ public class PolynomialApproximation {
if (debugLevel>3) System.out.println(">>> n="+n+" det_lin="+mLin.det()+" norm_lin="+normMatix(mAarrayL));
double nmL=normMatix(mAarrayL);
if ((nmL==0.0) || (Math.abs(mLin.det())/nmL<thresholdLin)){
// return average value for each channel
// return average value for each channel
if (S00==0.0) return null; // not even average
double [][] ABCDEF=new double[zDim][3];
for (i=0;i<zDim;i++) {
......@@ -540,7 +545,7 @@ public class PolynomialApproximation {
}
double []zAarrayL=new double [3];
double [][] ABCDEF=new double[zDim][];
// double [] zAarrayL={SZ10,SZ01,SZ00};
// double [] zAarrayL={SZ10,SZ01,SZ00};
for (i=0;i<zDim;i++) {
zAarrayL[0]=SZ10[i];
zAarrayL[1]=SZ01[i];
......@@ -575,7 +580,7 @@ public class PolynomialApproximation {
", Math.abs(M.det())/normMatix(mAarrayQ)="+(Math.abs(mQuad.det())/normMatix(mAarrayQ))); //did not happen
return ABCDEF; // not enough data for the quadratic approximation, return linear
}
// double [] zAarrayQ={SZ20,SZ02,SZ11,SZ10,SZ01,SZ00};
// double [] zAarrayQ={SZ20,SZ02,SZ11,SZ10,SZ01,SZ00};
double [] zAarrayQ=new double [6];
for (i=0;i<zDim;i++) {
zAarrayQ[0]=SZ20[i];
......@@ -589,7 +594,7 @@ public class PolynomialApproximation {
}
return ABCDEF;
}
// calculate "volume" made of the matrix row-vectors, placed orthogonally
// calculate "volume" made of the matrix row-vectors, placed orthogonally
// to be compared to determinant
public double normMatix(double [][] a) {
double d,norm=1.0;
......@@ -601,6 +606,161 @@ public class PolynomialApproximation {
return norm;
}
public static double [] getYXRegression(
final double [] data_x,
final double [] data_y,
final boolean [] mask) {
final Thread[] threads = ImageDtt.newThreadArray(QuadCLT.THREADS_MAX);
final AtomicInteger ai = new AtomicInteger(0);
final double [] as0 = new double[threads.length];
final double [] asx = new double[threads.length];
final double [] asx2 = new double[threads.length];
final double [] asy = new double[threads.length];
final double [] asxy= new double[threads.length];
final AtomicInteger ati = new AtomicInteger(0);
ai.set(0);
for (int ithread = 0; ithread < threads.length; ithread++) {
threads[ithread] = new Thread() {
public void run() {
int thread_num = ati.getAndIncrement();
for (int ipix = ai.getAndIncrement(); ipix < data_x.length; ipix = ai.getAndIncrement()) if ((mask == null) || mask[ipix]){
double x = data_x[ipix];
double y = data_y[ipix];
if (!Double.isNaN(x) && !Double.isNaN(y)) {
as0 [thread_num] += 1;
asx [thread_num] += x;
asx2[thread_num] += x * x;
asy [thread_num] += y;
asxy[thread_num] += x * y;
}
} // for (int ipix
}
};
}
ImageDtt.startAndJoin(threads);
double s0 = 0.0;
double sx = 0.0;
double sx2 = 0.0;
double sy = 0.0;
double sxy= 0.0;
for (int i = 0; i < threads.length; i++) {
s0+= as0[i];
sx+= asx[i];
sx2+= asx2[i];
sy+= asy[i];
sxy+= asxy[i];
}
double dnm = s0 * sx2 - sx*sx;
double a = (sxy * s0 - sy * sx) / dnm;
double b = (sy * sx2 - sxy * sx) / dnm;
return new double[] {a,b};
}
//RuntimeException
/**
* Get best fit ax+b symmetrical for X and Y, same weights
* https://en.wikipedia.org/wiki/Deming_regression#Orthogonal_regression
* @param data_x
* @param data_y
* @param mask
* @return
*/
public static double [] getOrthoRegression( // symmetrical for X and Y, same eror weights
final double [] data_x,
final double [] data_y,
final boolean [] mask_in) {
final boolean [] mask = new boolean [data_x.length];
final Thread[] threads = ImageDtt.newThreadArray(QuadCLT.THREADS_MAX);
final AtomicInteger ai = new AtomicInteger(0);
final double [] as0 = new double[threads.length];
final double [] asx = new double[threads.length];
final double [] asy = new double[threads.length];
final AtomicInteger ati = new AtomicInteger(0);
ai.set(0);
// Find centroid first
for (int ithread = 0; ithread < threads.length; ithread++) {
threads[ithread] = new Thread() {
public void run() {
int thread_num = ati.getAndIncrement();
for (int ipix = ai.getAndIncrement(); ipix < data_x.length; ipix = ai.getAndIncrement()) if ((mask_in == null) || mask_in[ipix]){
double x = data_x[ipix];
double y = data_y[ipix];
if (!Double.isNaN(x) && !Double.isNaN(y)) {
as0 [thread_num] += 1;
asx [thread_num] += x;
asy [thread_num] += y;
mask[ipix] = true;
}
} // for (int ipix
}
};
}
ImageDtt.startAndJoin(threads);
double s0 = 0.0;
double sx = 0.0;
double sy = 0.0;
for (int i = 0; i < threads.length; i++) {
s0+= as0[i];
sx+= asx[i];
sy+= asy[i];
}
double [] z_mean = {sx/s0, sy/s0}; // complex
final double [] as_re = new double[threads.length];
final double [] as_im = new double[threads.length];
ai.set(0);
ati.set(0);
for (int ithread = 0; ithread < threads.length; ithread++) {
threads[ithread] = new Thread() {
public void run() {
int thread_num = ati.getAndIncrement();
for (int ipix = ai.getAndIncrement(); ipix < data_x.length; ipix = ai.getAndIncrement()) if ((mask == null) || mask[ipix]){
double x = data_x[ipix]-z_mean[0];
double y = data_y[ipix]-z_mean[1];
as_re[thread_num] += x*x - y*y;
as_im[thread_num] += 2*x*y;
} // for (int ipix
}
};
}
ImageDtt.startAndJoin(threads);
double s_re = 0.0;
double s_im = 0.0;
for (int i = 0; i < threads.length; i++) {
s_re+= as_re[i];
s_im+= as_im[i];
}
// https://en.wikipedia.org/wiki/Square_root#Algebraic_formula
// sqrt (s_re+i*s_im)
double sqrt_re = Math.sqrt(0.5 * (Math.sqrt(s_re*s_re + s_im*s_im) +s_re));
double sqrt_im = ((s_im > 0)? 1 : -1) * Math.sqrt(0.5 * (Math.sqrt(s_re*s_re + s_im*s_im) - s_re));
double a = sqrt_im/sqrt_re;
double b = z_mean[1]- a * z_mean[0];
return new double[] {a,b};
}
public static void applyRegression(
final double [] data, // clone by caller
final double [] regression) {
final Thread[] threads = ImageDtt.newThreadArray(QuadCLT.THREADS_MAX);
final AtomicInteger ai = new AtomicInteger(0);
for (int ithread = 0; ithread < threads.length; ithread++) {
threads[ithread] = new Thread() {
public void run() {
for (int ipix = ai.getAndIncrement(); ipix < data.length; ipix = ai.getAndIncrement()){
data[ipix] = regression[0] * data[ipix] + regression[1];
}
}
};
}
ImageDtt.startAndJoin(threads);
}
public static double [] invertRegression(double [] regression) {
return new double [] {1.0/regression[0], -regression[1]/regression[0]};
}
}
src/main/java/com/elphel/imagej/orthomosaic/ComboMatch.java
View file @ 1671728b
......@@ -171,7 +171,7 @@ public class ComboMatch {
boolean pattern_match = true; // false;
boolean bounds_to_indices = true;
int temp_mode = 1;
int temp_mode = 0;
boolean restore_temp = true;
double frac_remove = clt_parameters.imp.pmtch_frac_remove; // 0.15;
double metric_error = clt_parameters.imp.pmtch_metric_err; // 0.05; // 0.02;// 2 cm
......@@ -1030,10 +1030,13 @@ public class ComboMatch {
}
if (render_match) {
String title=String.format("multi_%03d-%03d_%s-%s_zoom%d_%d",gpu_pair[0],gpu_pair[1],gpu_spair[0],gpu_spair[1],min_zoom_lev,zoom_lev);
// Avoid renderMulti() - it duplicates code renderMultiDouble()
int eq_mode = 2; // calculate
ImagePlus imp_img_pair = maps_collection.renderMulti (
//_zoom<integer> is needed for opening with "Extract Objects" command
title, // String title,
OrthoMapsCollection.MODE_IMAGE, // int mode, // 0 - regular image, 1 - altitudes, 2 - black/white mask // boolean use_alt,
// OrthoMapsCollection.MODE_IMAGE, // int mode, // 0 - regular image, 1 - altitudes, 2 - black/white mask // boolean use_alt,
eq_mode, //int eq_mode, // 0 - ignore equalization, 1 - use stored equalization, 2 - calculate equalization
gpu_pair, // int [] indices, // null or which indices to use (normally just 2 for pairwise comparison)
bounds_to_indices, // boolean bounds_to_indices,
temp_mode, // int temp_mode, // 0 - do nothing, 1 - equalize average,2 - try to correct
......
src/main/java/com/elphel/imagej/orthomosaic/OrthoMap.java
View file @ 1671728b
......@@ -115,6 +115,7 @@ public class OrthoMap implements Comparable <OrthoMap>, Serializable{
public transient double agl = Double.NaN;
public transient int num_scenes = -1;; // number of scenes that made up this image
public transient double sfm_gain = Double.NaN; // maximal SfM gain of this map
public transient double [] equalize = {1,0}; // rectified value = equalize[0]*source_value+equalize[1]
private void writeObject(ObjectOutputStream oos) throws IOException {
oos.defaultWriteObject();
oos.writeObject(path);
......@@ -137,6 +138,7 @@ public class OrthoMap implements Comparable <OrthoMap>, Serializable{
oos.writeObject(agl);
oos.writeObject(num_scenes);
oos.writeObject(sfm_gain);
oos.writeObject(equalize);
}
private void readObject(ObjectInputStream ois) throws ClassNotFoundException, IOException {
......@@ -160,6 +162,8 @@ public class OrthoMap implements Comparable <OrthoMap>, Serializable{
agl = (double) ois.readObject();
num_scenes = (int) ois.readObject();
sfm_gain = (double) ois.readObject();
// equalize = new double[] {1,0};
equalize = (double []) ois.readObject();
images = new HashMap <Integer, FloatImageData>(); // field images was not saved
averageImagePixel = Double.NaN; // average image pixel value (to combine with raw)
......@@ -167,6 +171,18 @@ public class OrthoMap implements Comparable <OrthoMap>, Serializable{
// pairwise_matches = new HashMap<Double, PairwiseOrthoMatch>();
}
double getEqualized(double d) {
return d * equalize[0] + equalize[1];
}
double [] getEqualize() {
return equalize;
}
void setEqualize(double [] equalize) {
this.equalize = equalize;
}
@Override
public int compareTo(OrthoMap otherPlayer) {
return Double.compare(ts, otherPlayer.ts);
......@@ -1717,9 +1733,9 @@ public class OrthoMap implements Comparable <OrthoMap>, Serializable{
};
}
ImageDtt.startAndJoin(threads);
double [] ab = getDatiRegression(
temp, // final double [] temp,
dati, // final double [] dati,
double [] ab = PolynomialApproximation.getYXRegression(
temp, // final double [] data_x,
dati, // final double [] data_y,
flat); // final boolean [] mask);
double a = ab[0];
double b = ab[1];
......@@ -1774,56 +1790,6 @@ public class OrthoMap implements Comparable <OrthoMap>, Serializable{
return flat;
}
private static double [] getDatiRegression(
final double [] temp,
final double [] dati,
final boolean [] mask) {
final Thread[] threads = ImageDtt.newThreadArray(QuadCLT.THREADS_MAX);
final AtomicInteger ai = new AtomicInteger(0);
final double [] as0 = new double[threads.length];
final double [] asx = new double[threads.length];
final double [] asx2 = new double[threads.length];
final double [] asy = new double[threads.length];
final double [] asxy= new double[threads.length];
final AtomicInteger ati = new AtomicInteger(0);
ai.set(0);
for (int ithread = 0; ithread < threads.length; ithread++) {
threads[ithread] = new Thread() {
public void run() {
int thread_num = ati.getAndIncrement();
for (int ipix = ai.getAndIncrement(); ipix < temp.length; ipix = ai.getAndIncrement()) if (mask[ipix]){
double x = temp[ipix];
double y = dati[ipix];
if (!Double.isNaN(x+y)) {
as0 [thread_num] += 1;
asx [thread_num] += x;
asx2[thread_num] += x * x;
asy [thread_num] += y;
asxy[thread_num] += x * y;
}
} // for (int ipix
}
};
}
ImageDtt.startAndJoin(threads);
double s0 = 0.0;
double sx = 0.0;
double sx2 = 0.0;
double sy = 0.0;
double sxy= 0.0;
for (int i = 0; i < threads.length; i++) {
s0+= as0[i];
sx+= asx[i];
sx2+= asx2[i];
sy+= asy[i];
sxy+= asxy[i];
}
double dnm = s0 * sx2 - sx*sx;
double a = (sxy * s0 - sy * sx) / dnm;
double b = (sy * sx2 - sxy * sx) / dnm;
return new double[] {a,b};
}
private static int [][] getCirclePoints(
double radius) {
......
src/main/java/com/elphel/imagej/orthomosaic/OrthoMultiLMA.java
View file @ 1671728b
......@@ -41,6 +41,7 @@ import ij.gui.PointRoi;
public class OrthoMultiLMA {
final boolean move_only;
final boolean corr_avg; // correct average skew, tilt, scale
private double [] last_rms = null; // {rms, rms_pure}, matching this.vector
private double [] good_or_bad_rms = null; // just for diagnostics, to read last (failed) rms
private double [] initial_rms = null; // {rms, rms_pure}, first-calcualted rms
......@@ -61,7 +62,11 @@ public class OrthoMultiLMA {
private double [][] offsets = null; // scene offsets (rd)
private int num_scenes = 0;
private int num_pairs = 0;
public OrthoMultiLMA (boolean move_only) {
public OrthoMultiLMA (
boolean corr_avg,
boolean move_only
) {
this.corr_avg = corr_avg && !move_only;
this.move_only = move_only;
}
public static boolean testMultiLMA(
......@@ -122,6 +127,8 @@ public class OrthoMultiLMA {
indices, // int [] indices, // null or which indices to use (normally just 2 for pairwise comparison)
true, // boolean bounds_to_indices,
affines_a, // affines_0d, // affines, // double [][][] affines, // null or [indices.length][2][3]
null, // double [][] equalize,
true, // boolean ignore_equalize,
null, // warp, // FineXYCorr warp,,
zoom_level, // int zoom_level,
wh, // int [] wh,
......@@ -147,6 +154,8 @@ public class OrthoMultiLMA {
indices, // int [] indices, // null or which indices to use (normally just 2 for pairwise comparison)
true, // boolean bounds_to_indices,
affines_sym,// affines, // double [][][] affines, // null or [indices.length][2][3]
null, // double [][] equalize,
true, // boolean ignore_equalize,
null, // warp, // FineXYCorr warp,,
zoom_level, // int zoom_level,
wh, // int [] wh,
......@@ -213,19 +222,26 @@ public class OrthoMultiLMA {
int debugLevel = 1;
boolean move_only = false;
double [] val_coord = null; // 1 - valid, 0 - invalid, minimize coordinates errors
double position_pull = 0.001;
boolean use_inv = false;
double overlap_pow = 2.0; // match weight as overlap fraction to this power
double skew_pull = 1.0;
double tilt_pull = 1.0;
double scale_pull = 0.1; // .0;
double position_pull = 0.0001;
boolean corr_avg= (skew_pull > 0) || (tilt_pull > 0) || (scale_pull > 0);
int [] indices = maps_collection.getScenesSelection(
null, // boolean select_all,
" to build a map"); // String purpose)
OrthoMultiLMA oml = new OrthoMultiLMA(move_only);
OrthoMultiLMA oml = new OrthoMultiLMA(
corr_avg,
move_only);
double lambda = 0.1;
double lambda_scale_good = 0.5;
double lambda_scale_bad = 8.0;
double lambda_max = 100;
double rms_diff = 0.000001;
int num_iter = 20;
int num_iter = 100; // 50;
boolean last_run = false;
oml.prepareLMA (
clt_parameters, // CLTParameters clt_parameters,
......@@ -233,6 +249,9 @@ public class OrthoMultiLMA {
indices, // int [] indices,
val_coord, // double [] val_coord, // 1 - valid, 0 - invalid, minimize coordinates errors
position_pull, // double position_pull,
skew_pull, // double skew_pull,
tilt_pull, // double tilt_pull,
scale_pull, // double scale_pull,
use_inv, // boolean use_inv,
overlap_pow, // double overlap_pow, // match weight as overlap fraction to this power
debugLevel); // int debugLevel)
......@@ -256,7 +275,7 @@ public class OrthoMultiLMA {
//Get and apply affines
//oml.updateAffines(maps_collection);
double [][][] affines = oml.getAffines();
double [] fx = oml.getFx();
// test
/*
PairwiseOrthoMatch match = maps_collection.ortho_maps[indices[0]].getMatch(maps_collection.ortho_maps[indices[1]].getName());
......@@ -284,6 +303,8 @@ public class OrthoMultiLMA {
indices, // int [] indices, // null or which indices to use (normally just 2 for pairwise comparison)
true, // boolean bounds_to_indices,
affines, // null, // affines, // double [][][] affines, // null or [indices.length][2][3]
null, // double [][] equalize,
true, // boolean ignore_equalize,
null, // warp, // FineXYCorr warp,,
zoom_level, // int zoom_level,
wh, // int [] wh,
......@@ -307,6 +328,7 @@ public class OrthoMultiLMA {
roi.setOptions("label");
imp_multi.setRoi(roi);
imp_multi.show();
/*
double [][] affine = getAffineAndDerivatives(
move_only, //boolean move_only,
affines[0], // double [][] affine00,
......@@ -317,7 +339,7 @@ public class OrthoMultiLMA {
affines[0],
affines[1],
oml.offsets[0]);
*/
/*
......@@ -336,6 +358,16 @@ public class OrthoMultiLMA {
return 0;
}
public double [] getParameters() {
return parameters_vector;
}
public double [] getFx() {
return getFxDerivs(
parameters_vector, // double [] vector,
null, // final double [][] jt, // should be null or initialized with [vector.length][]
0); // final int debug_level)
}
public int prepareLMA (
CLTParameters clt_parameters,
......@@ -343,6 +375,9 @@ public class OrthoMultiLMA {
int [] indices,
double [] val_coord, // 1 - valid, 0 - invalid, minimize coordinates errors
double position_pull,
double skew_pull,
double tilt_pull,
double scale_pull,
boolean use_inv,
double overlap_pow, // match weight as overlap fraction to this power
int debugLevel) {
......@@ -383,7 +418,6 @@ public class OrthoMultiLMA {
for (int np = 0; np < num_pairs; np++) {
Point pair = pairs_list.get(np);
pairs[np] = new int[] {pair.x, pair.y};
// double [] enuOffset = maps_collection.ortho_maps[indices[pair.x]].enuOffsetTo(maps_collection.ortho_maps[indices[pair.y]]);
double [] enuOffset = maps_collection.ortho_maps[indices[pair.y]].enuOffsetTo(maps_collection.ortho_maps[indices[pair.x]]);
offsets[np] = new double[] {enuOffset[0], -enuOffset[1]}; // {right,down} of the image
}
......@@ -396,9 +430,8 @@ public class OrthoMultiLMA {
int n62 = move_only ? 2 : 6;
int n13 = move_only ? 1 : 3;
y_vector = new double [n62*num_pairs + 2 * num_scenes]; // last 2 * num_scenes will stay 0
weights = new double [n62*num_pairs + 2 * num_scenes];
y_vector = new double [n62*num_pairs + 2 * num_scenes + (corr_avg ? 3:0)]; // last 2 * num_scenes + 3 will stay 0
weights = new double [n62*num_pairs + 2 * num_scenes + (corr_avg ? 3:0)];
parameters_vector = new double [n62*num_scenes]; // maybe will need move-only mode?
for (int n = 0; n < num_scenes; n++) {
double [][] affine = maps_collection.ortho_maps[indices[n]].getAffine();
......@@ -442,6 +475,11 @@ public class OrthoMultiLMA {
weights[wi++] = w;
sw+= 2*w;
}
if (corr_avg) {
weights[wi++] = skew_pull;
weights[wi++] = tilt_pull;
weights[wi++] = scale_pull;
}
pure_weight = swp/sw;
double s = 1.0/sw;
for (int i = 0; i <weights.length;i++) {
......@@ -1166,6 +1204,49 @@ public class OrthoMultiLMA {
};
}
ImageDtt.startAndJoin(threads);
ai.set(0);
if (corr_avg) { // && ((skew_pull > 0) || (tilt_pull > 0) || (scale_pull > 0))) {
final int pull_indices = weights.length-3;
for (int nScene = 0; nScene < num_scenes; nScene++){
int indx = 6 * nScene;
double a00 = vector[indx+0],a01=vector[indx+1],a10=vector[indx+3],a11=vector[indx+4];
double aa1 = 0.5*(a00*a00 + a10*a10);
double aa2 = 0.5*(a01*a01 + a11*a11);
fx[pull_indices + 0] += a00*a01+a10*a11;
fx[pull_indices + 1] += aa1-aa2;
fx[pull_indices + 2] += aa1+aa2 - 1.0;
}
if (jt!= null) {
for (int ithread = 0; ithread < threads.length; ithread++) {
threads[ithread] = new Thread() {
public void run() {
for (int nScene = ai.getAndIncrement(); nScene < num_scenes; nScene = ai.getAndIncrement()){
int indx = 6 * nScene;
double a00=vector[indx+0],a01=vector[indx+1],a10=vector[indx+3],a11=vector[indx+4];
jt[indx + 0][pull_indices + 0] = a01;
jt[indx + 1][pull_indices + 0] = a00;
jt[indx + 3][pull_indices + 0] = a11;
jt[indx + 4][pull_indices + 0] = a10;
jt[indx + 0][pull_indices + 1] = a00;
jt[indx + 1][pull_indices + 1] =-a01;
jt[indx + 3][pull_indices + 1] = a10;
jt[indx + 4][pull_indices + 1] =-a11;
jt[indx + 0][pull_indices + 2] = a00;
jt[indx + 1][pull_indices + 2] = a01;
jt[indx + 3][pull_indices + 2] = a10;
jt[indx + 4][pull_indices + 2] = a11;
}
}
};
}
ImageDtt.startAndJoin(threads);
ai.set(0);
}
}
return fx;
}
......
src/main/java/com/elphel/imagej/orthomosaic/PairwiseOrthoMatch.java
View file @ 1671728b
......@@ -15,6 +15,8 @@ public class PairwiseOrthoMatch implements Serializable {
public double rms = Double.NaN;
public transient int [] nxy = null; // not saved, just to communicate for logging
public transient double overlap = 0.0;
public transient double [] equalize1to0 = {1,0}; // value1 = equalize2to1[0]*value2+equalize2to1[1]
public PairwiseOrthoMatch() {
}
......@@ -30,6 +32,11 @@ public class PairwiseOrthoMatch implements Serializable {
this.rms = rms;
this.overlap = overlap;
}
public void setEqualize2to1 (double [] equalize2to1) {
this.equalize1to0 = equalize2to1;
}
public PairwiseOrthoMatch clone() {
double [][] affine = {this.affine[0].clone(),this.affine[1].clone()};
double [][] jtj = new double [this.jtj.length][];
......@@ -45,6 +52,7 @@ public class PairwiseOrthoMatch implements Serializable {
if (nxy != null) {
pom.nxy = nxy.clone();
}
pom.equalize1to0 = this.equalize1to0.clone();
return pom;
}
/**
......@@ -102,6 +110,7 @@ public class PairwiseOrthoMatch implements Serializable {
rd[0] * affine[1][0]+ rd[1]*(affine[1][1]-1.0)};
affine[0][2] += corr[0];
affine[1][2] += corr[1];
inverted_match.setEqualize2to1(new double [] {1/equalize1to0[0], -equalize1to0[1]/equalize1to0[0]});
return inverted_match;
}
......@@ -168,10 +177,16 @@ public class PairwiseOrthoMatch implements Serializable {
affine = new double[][] {
{A.get(0,0),A.get(0,1), B.get(0,0)},
{A.get(1,0),A.get(1,1), B.get(1,0)}};
// jtj = null;
// rms = Double.NaN; // double rms,
// zoom_lev = 0; // int zoom_lev)
}
public void combineEqualize(
double [] equalize0,
double [] equalize1 ) {
setEqualize2to1(new double[] {
equalize1[0]/equalize0[0],
equalize1[1]-equalize1[0]/equalize0[0]*equalize0[1]});
}
public static double [][] combineAffines(
double [][] affine0,
double [][] affine, // differential
......@@ -203,6 +218,8 @@ public class PairwiseOrthoMatch implements Serializable {
}
}
oos.writeObject(overlap);
oos.writeObject(equalize1to0);
}
private void readObject(ObjectInputStream ois) throws ClassNotFoundException, IOException {
ois.defaultReadObject();
......@@ -216,6 +233,8 @@ public class PairwiseOrthoMatch implements Serializable {
}
}
overlap = (Double) ois.readObject();
// equalize1to0 = new double[] {1,0};
equalize1to0 = (double[]) ois.readObject();
}
//private void readObjectNoData() throws ObjectStreamException; // used to modify default values
}
src/main/java/com/elphel/imagej/tileprocessor/TileNeibs.java
View file @ 1671728b
......@@ -1356,4 +1356,5 @@ public class TileNeibs{
}
return filled;
}
}
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment