// returns value of mean intensity+3*SD based on
// fitting of image histogram to gaussian function
int getThreshold(ImageProcessor thImage) {
ImageStatistics imgstat;
double[][] dHistogram;
double[] dHistCum;
double[][] dNoiseFit;
int nHistSize;
int nCount, nMaxCount;
int nDownCount, nUpCount;
int i, nPeakPos;
double dRightWidth, dLeftWidth, dWidth;
double dMean, dSD;
double[] dFitErrors;
double dErrCoeff;
double dSum = 0.0;
double dSumFit = 0.0;
LMA fitlma;
// mean, sd, min, max
// imgstat = ImageStatistics.getStatistics(thImage, 38, null);
imgstat =
ImageStatistics.getStatistics(
thImage, Measurements.MEAN + Measurements.STD_DEV + Measurements.MIN_MAX, null);
dMean = imgstat.mean;
nPeakPos = 0;
nHistSize = imgstat.histogram.length;
dHistogram = new double[2][nHistSize];
nMaxCount = 0;
// determine position and height of maximum count in histogram (mode)
// and height at maximum
for (i = 0; i < nHistSize; i++) {
nCount = imgstat.histogram[i];
dHistogram[0][i] = imgstat.min + i * imgstat.binSize;
dHistogram[1][i] = (double) nCount;
dSum += (double) nCount;
if (nMaxCount < nCount) {
nMaxCount = nCount;
dMean = imgstat.min + i * imgstat.binSize;
nPeakPos = i;
}
}
// estimating width of a peak
// going to the left
i = nPeakPos;
while (i > 0 && imgstat.histogram[i] > 0.5 * nMaxCount) {
i--;
}
if (i < 0) i = 0;
dLeftWidth = i;
// going to the right
i = nPeakPos;
while (i < nHistSize && imgstat.histogram[i] > 0.5 * nMaxCount) {
i++;
}
if (i == nHistSize) i = nHistSize - 1;
dRightWidth = i;
// FWHM in bins
dWidth = (dRightWidth - dLeftWidth);
dSD = dWidth * imgstat.binSize / 2.35;
// fitting range +/- 3*SD
dLeftWidth = nPeakPos - 3 * dWidth / 2.35;
if (dLeftWidth < 0) dLeftWidth = 0;
dRightWidth = nPeakPos + 3 * dWidth / 2.35;
if (dRightWidth > nHistSize) dRightWidth = nHistSize;
nUpCount = (int) dRightWidth;
nDownCount = (int) dLeftWidth;
// preparing histogram range for fitting
dNoiseFit = new double[2][nUpCount - nDownCount + 1];
for (i = nDownCount; i <= nUpCount; i++) {
dNoiseFit[0][i - nDownCount] = dHistogram[0][i];
dNoiseFit[1][i - nDownCount] = dHistogram[1][i];
dSumFit += dHistogram[1][i];
}
// fitting range is too small, less than 80%
if ((dSumFit / dSum) < 0.8) {
// build cumulative distribution
dHistCum = new double[nHistSize];
dSumFit = 0;
for (i = 0; i < nHistSize; i++) {
dSumFit += dHistogram[1][i];
dHistCum[i] = dSumFit / dSum;
}
nUpCount = 0;
nDownCount = 0;
for (i = 0; i < nHistSize; i++) {
// 10% quantile
if (dHistCum[i] < 0.11) nDownCount = i;
// 90% quantile
if (dHistCum[i] < 0.91) nUpCount = i;
}
// preparing histogram range for fitting
dNoiseFit = new double[2][nUpCount - nDownCount + 1];
for (i = nDownCount; i <= nUpCount; i++) {
dNoiseFit[0][i - nDownCount] = dHistogram[0][i];
dNoiseFit[1][i - nDownCount] = dHistogram[1][i];
}
}
fitlma =
new LMA(new SMLOneDGaussian(), new double[] {(double) nMaxCount, dMean, dSD}, dNoiseFit);
fitlma.fit();
dMean = fitlma.parameters[1];
dSD = fitlma.parameters[2];
dFitErrors = fitlma.getStandardErrorsOfParameters();
// scaling coefficient for parameters errors estimation
// (Standard deviation of residuals)
dErrCoeff = Math.sqrt(fitlma.chi2 / (nUpCount - nDownCount + 1 - 3));
for (i = 0; i < 3; i++) dFitErrors[i] *= dErrCoeff;
for (i = 0; i < 3; i++) dFitErrors[i] *= 100 / fitlma.parameters[i];
if (dFitErrors[1] > 20
|| dMean < imgstat.min
|| dMean > imgstat.max
|| dSD < imgstat.min
|| dSD > imgstat.max)
// fit somehow failed
return (int) (imgstat.mean + 3.0 * imgstat.stdDev);
else return (int) (dMean + 3.0 * dSD);
}