@Test
public void testProcessNoNoise() {
int width = 100;
int height = 100;
final int nspots = 4;
long[] dimensions = new long[] {width, height};
ArrayImg<UnsignedByteType, ByteArray> img = ArrayImgs.unsignedBytes(dimensions);
final double[] A0 = new double[nspots];
final double[] x0 = new double[nspots];
final double[] y0 = new double[nspots];
final double[] sigma_x = new double[nspots];
final double[] sigma_y = new double[nspots];
Random rangen = new Random(1);
for (int i = 0; i < nspots; i++) {
A0[i] = 100 + 10 * rangen.nextGaussian();
x0[i] = 20 + (width - 20) * rangen.nextDouble();
y0[i] = 20 + (height - 20) * rangen.nextDouble();
sigma_x[i] = 2 + 0.5 * rangen.nextGaussian();
sigma_y[i] = 2 + 0.5 * rangen.nextGaussian();
double[] params =
new double[] {
A0[i], x0[i], y0[i], 1 / sigma_x[i] / sigma_x[i], 1 / sigma_y[i] / sigma_y[i]
};
LocalizationUtils.addGaussianSpotToImage(img, params);
}
// Instantiate fitter once
final GaussianPeakFitterND<UnsignedByteType> fitter =
new GaussianPeakFitterND<UnsignedByteType>(img);
// No multithreading in the test
for (int j = 0; j < nspots; j++) {
Localizable point = new Point(new int[] {(int) x0[j], (int) y0[j]});
double[] results = fitter.process(point, new double[] {sigma_x[j], sigma_y[j]});
double A = results[0];
double x = results[1];
double y = results[2];
double sx = 1 / Math.sqrt(results[3]);
double sy = 1 / Math.sqrt(results[4]);
assertEq("x", x0[j], x, LOCALIZATION_TOLERANCE_NO_NOISE);
assertEq("y", y0[j], y, LOCALIZATION_TOLERANCE_NO_NOISE);
assertEq("σx", sigma_x[j], sx, SIGMA_TOLERANCE_NO_NOISE);
assertEq("σy", sigma_y[j], sy, SIGMA_TOLERANCE_NO_NOISE);
assertEq("A", A0[j], A, 20d);
}
}
@Test
public void testSymetricGaussian() {
int width = 200;
int height = 200;
final int nspots = 10;
long[] dimensions = new long[] {width, height};
ArrayImg<UnsignedByteType, ByteArray> img = ArrayImgs.unsignedBytes(dimensions);
Collection<Localizable> peaks = new HashSet<Localizable>(nspots);
Map<Localizable, double[]> groundTruth = new HashMap<Localizable, double[]>(nspots);
for (int i = 1; i < nspots; i++) {
for (int j = 1; j < nspots; j++) {
double A = 100;
double x0 = width / (double) nspots * i * 1.02d;
double y0 = width / (double) nspots * j * 1.02d;
double sigma = 2 - (double) i / nspots + (double) i / nspots;
Localizable peak = new Point((long) x0, (long) y0);
peaks.add(peak);
double[] params = new double[] {x0, y0, A, 1 / sigma / sigma};
LocalizationUtils.addGaussianSpotToImage(img, params);
groundTruth.put(peak, params);
}
}
// Instantiate fitter once
PeakFitter<UnsignedByteType> fitter =
new PeakFitter<UnsignedByteType>(
img,
peaks,
new LevenbergMarquardtSolver(),
new Gaussian(),
new MLGaussianEstimator(2d, 2));
if (!fitter.checkInput() || !fitter.process()) {
fail("Problem with peak fitting: " + fitter.getErrorMessage());
return;
}
Map<Localizable, double[]> results = fitter.getResult();
for (Localizable peak : peaks) {
double[] params = results.get(peak);
double[] truth = groundTruth.get(peak);
// Pedestrian assertion
assertEquals(
"Bad accuracy on amplitude parameter A: ", truth[2], params[2], TOLERANCE * truth[2]);
assertEquals(
"Bad accuracy on peak location x0: ", truth[0], params[0], LOCALIZATION_TOLERANCE);
assertEquals(
"Bad accuracy on peak location y0: ", truth[1], params[1], LOCALIZATION_TOLERANCE);
assertEquals("Bad accuracy on peak paramter b: ", truth[3], params[3], TOLERANCE * truth[3]);
}
}