/** Generate output image whose type is same as input image. */
private ImagePlus makeOutputImage(ImagePlus imp, FloatProcessor fp, int ptype) {
int width = imp.getWidth();
int height = imp.getHeight();
float[] pixels = (float[]) fp.getPixels();
ImageProcessor oip = null;
// Create output image consistent w/ type of input image.
int size = pixels.length;
switch (ptype) {
case BYTE_TYPE:
oip = imp.getProcessor().createProcessor(width, height);
byte[] pixels8 = (byte[]) oip.getPixels();
for (int i = 0; i < size; i++) pixels8[i] = (byte) pixels[i];
break;
case SHORT_TYPE:
oip = imp.getProcessor().createProcessor(width, height);
short[] pixels16 = (short[]) oip.getPixels();
for (int i = 0; i < size; i++) pixels16[i] = (short) pixels[i];
break;
case FLOAT_TYPE:
oip = new FloatProcessor(width, height, pixels, null);
break;
}
// Adjust for display.
// Calling this on non-ByteProcessors ensures image
// processor is set up to correctly display image.
oip.resetMinAndMax();
// Create new image plus object. Don't use
// ImagePlus.createImagePlus here because there may be
// attributes of input image that are not appropriate for
// projection.
return new ImagePlus(makeTitle(), oip);
}
void Bernsen(ImagePlus imp, int radius, double par1, double par2, boolean doIwhite) {
// Bernsen recommends WIN_SIZE = 31 and CONTRAST_THRESHOLD = 15.
// 1) Bernsen J. (1986) "Dynamic Thresholding of Grey-Level Images"
// Proc. of the 8th Int. Conf. on Pattern Recognition, pp. 1251-1255
// 2) Sezgin M. and Sankur B. (2004) "Survey over Image Thresholding
// Techniques and Quantitative Performance Evaluation" Journal of
// Electronic Imaging, 13(1): 146-165
// http://citeseer.ist.psu.edu/sezgin04survey.html
// Ported to ImageJ plugin from E Celebi's fourier_0.8 routines
// This version uses a circular local window, instead of a rectagular one
ImagePlus Maximp, Minimp;
ImageProcessor ip = imp.getProcessor(), ipMax, ipMin;
int contrast_threshold = 15;
int local_contrast;
int mid_gray;
byte object;
byte backg;
int temp;
if (par1 != 0) {
IJ.log("Bernsen: changed contrast_threshold from :" + contrast_threshold + " to:" + par1);
contrast_threshold = (int) par1;
}
if (doIwhite) {
object = (byte) 0xff;
backg = (byte) 0;
} else {
object = (byte) 0;
backg = (byte) 0xff;
}
Maximp = duplicateImage(ip);
ipMax = Maximp.getProcessor();
RankFilters rf = new RankFilters();
rf.rank(ipMax, radius, rf.MAX); // Maximum
// Maximp.show();
Minimp = duplicateImage(ip);
ipMin = Minimp.getProcessor();
rf.rank(ipMin, radius, rf.MIN); // Minimum
// Minimp.show();
byte[] pixels = (byte[]) ip.getPixels();
byte[] max = (byte[]) ipMax.getPixels();
byte[] min = (byte[]) ipMin.getPixels();
for (int i = 0; i < pixels.length; i++) {
local_contrast = (int) ((max[i] & 0xff) - (min[i] & 0xff));
mid_gray = (int) ((min[i] & 0xff) + (max[i] & 0xff)) / 2;
temp = (int) (pixels[i] & 0x0000ff);
if (local_contrast < contrast_threshold)
pixels[i] = (mid_gray >= 128) ? object : backg; // Low contrast region
else pixels[i] = (temp >= mid_gray) ? object : backg;
}
// imp.updateAndDraw();
return;
}
void Contrast(ImagePlus imp, int radius, double par1, double par2, boolean doIwhite) {
// G. Landini, 2013
// Based on a simple contrast toggle. This procedure does not have user-provided paramters other
// than the kernel radius
// Sets the pixel value to either white or black depending on whether its current value is
// closest to the local Max or Min respectively
// The procedure is similar to Toggle Contrast Enhancement (see Soille, Morphological Image
// Analysis (2004), p. 259
ImagePlus Maximp, Minimp;
ImageProcessor ip = imp.getProcessor(), ipMax, ipMin;
int c_value = 0;
int mid_gray;
byte object;
byte backg;
if (doIwhite) {
object = (byte) 0xff;
backg = (byte) 0;
} else {
object = (byte) 0;
backg = (byte) 0xff;
}
Maximp = duplicateImage(ip);
ipMax = Maximp.getProcessor();
RankFilters rf = new RankFilters();
rf.rank(ipMax, radius, rf.MAX); // Maximum
// Maximp.show();
Minimp = duplicateImage(ip);
ipMin = Minimp.getProcessor();
rf.rank(ipMin, radius, rf.MIN); // Minimum
// Minimp.show();
byte[] pixels = (byte[]) ip.getPixels();
byte[] max = (byte[]) ipMax.getPixels();
byte[] min = (byte[]) ipMin.getPixels();
for (int i = 0; i < pixels.length; i++) {
pixels[i] =
((Math.abs((int) (max[i] & 0xff - pixels[i] & 0xff))
<= Math.abs((int) (pixels[i] & 0xff - min[i] & 0xff))))
? object
: backg;
}
// imp.updateAndDraw();
return;
}
void MidGrey(ImagePlus imp, int radius, double par1, double par2, boolean doIwhite) {
// See: Image Processing Learning Resourches HIPR2
// http://homepages.inf.ed.ac.uk/rbf/HIPR2/adpthrsh.htm
ImagePlus Maximp, Minimp;
ImageProcessor ip = imp.getProcessor(), ipMax, ipMin;
int c_value = 0;
int mid_gray;
byte object;
byte backg;
if (par1 != 0) {
IJ.log("MidGrey: changed c_value from :" + c_value + " to:" + par1);
c_value = (int) par1;
}
if (doIwhite) {
object = (byte) 0xff;
backg = (byte) 0;
} else {
object = (byte) 0;
backg = (byte) 0xff;
}
Maximp = duplicateImage(ip);
ipMax = Maximp.getProcessor();
RankFilters rf = new RankFilters();
rf.rank(ipMax, radius, rf.MAX); // Maximum
// Maximp.show();
Minimp = duplicateImage(ip);
ipMin = Minimp.getProcessor();
rf.rank(ipMin, radius, rf.MIN); // Minimum
// Minimp.show();
byte[] pixels = (byte[]) ip.getPixels();
byte[] max = (byte[]) ipMax.getPixels();
byte[] min = (byte[]) ipMin.getPixels();
for (int i = 0; i < pixels.length; i++) {
pixels[i] =
((int) (pixels[i] & 0xff) > (int) (((max[i] & 0xff) + (min[i] & 0xff)) / 2) - c_value)
? object
: backg;
}
// imp.updateAndDraw();
return;
}
void Mean(ImagePlus imp, int radius, double par1, double par2, boolean doIwhite) {
// See: Image Processing Learning Resourches HIPR2
// http://homepages.inf.ed.ac.uk/rbf/HIPR2/adpthrsh.htm
ImagePlus Meanimp;
ImageProcessor ip = imp.getProcessor(), ipMean;
int c_value = 0;
byte object;
byte backg;
if (par1 != 0) {
IJ.log("Mean: changed c_value from :" + c_value + " to:" + par1);
c_value = (int) par1;
}
if (doIwhite) {
object = (byte) 0xff;
backg = (byte) 0;
} else {
object = (byte) 0;
backg = (byte) 0xff;
}
Meanimp = duplicateImage(ip);
ImageConverter ic = new ImageConverter(Meanimp);
ic.convertToGray32();
ipMean = Meanimp.getProcessor();
RankFilters rf = new RankFilters();
rf.rank(ipMean, radius, rf.MEAN); // Mean
// Meanimp.show();
byte[] pixels = (byte[]) ip.getPixels();
float[] mean = (float[]) ipMean.getPixels();
for (int i = 0; i < pixels.length; i++)
pixels[i] = ((int) (pixels[i] & 0xff) > (int) (mean[i] - c_value)) ? object : backg;
// imp.updateAndDraw();
return;
}
void Sauvola(ImagePlus imp, int radius, double par1, double par2, boolean doIwhite) {
// Sauvola recommends K_VALUE = 0.5 and R_VALUE = 128.
// This is a modification of Niblack's thresholding method.
// Sauvola J. and Pietaksinen M. (2000) "Adaptive Document Image Binarization"
// Pattern Recognition, 33(2): 225-236
// http://www.ee.oulu.fi/mvg/publications/show_pdf.php?ID=24
// Ported to ImageJ plugin from E Celebi's fourier_0.8 routines
// This version uses a circular local window, instead of a rectagular one
ImagePlus Meanimp, Varimp;
ImageProcessor ip = imp.getProcessor(), ipMean, ipVar;
double k_value = 0.5;
double r_value = 128;
byte object;
byte backg;
if (par1 != 0) {
IJ.log("Sauvola: changed k_value from :" + k_value + " to:" + par1);
k_value = par1;
}
if (par2 != 0) {
IJ.log("Sauvola: changed r_value from :" + r_value + " to:" + par2);
r_value = par2;
}
if (doIwhite) {
object = (byte) 0xff;
backg = (byte) 0;
} else {
object = (byte) 0;
backg = (byte) 0xff;
}
Meanimp = duplicateImage(ip);
ImageConverter ic = new ImageConverter(Meanimp);
ic.convertToGray32();
ipMean = Meanimp.getProcessor();
RankFilters rf = new RankFilters();
rf.rank(ipMean, radius, rf.MEAN); // Mean
// Meanimp.show();
Varimp = duplicateImage(ip);
ic = new ImageConverter(Varimp);
ic.convertToGray32();
ipVar = Varimp.getProcessor();
rf.rank(ipVar, radius, rf.VARIANCE); // Variance
// Varimp.show();
byte[] pixels = (byte[]) ip.getPixels();
float[] mean = (float[]) ipMean.getPixels();
float[] var = (float[]) ipVar.getPixels();
for (int i = 0; i < pixels.length; i++)
pixels[i] =
((int) (pixels[i] & 0xff)
> (int) (mean[i] * (1.0 + k_value * ((Math.sqrt(var[i]) / r_value) - 1.0))))
? object
: backg;
// imp.updateAndDraw();
return;
}
void Phansalkar(ImagePlus imp, int radius, double par1, double par2, boolean doIwhite) {
// This is a modification of Sauvola's thresholding method to deal with low contrast images.
// Phansalskar N. et al. Adaptive local thresholding for detection of nuclei in diversity
// stained
// cytology images.International Conference on Communications and Signal Processing (ICCSP),
// 2011,
// 218 - 220.
// In this method, the threshold t = mean*(1+p*exp(-q*mean)+k*((stdev/r)-1))
// Phansalkar recommends k = 0.25, r = 0.5, p = 2 and q = 10. In this plugin, k and r are the
// parameters 1 and 2 respectively, but the values of p and q are fixed.
//
// Implemented from Phansalkar's paper description by G. Landini
// This version uses a circular local window, instead of a rectagular one
ImagePlus Meanimp, Varimp, Orimp;
ImageProcessor ip = imp.getProcessor(), ipMean, ipVar, ipOri;
double k_value = 0.25;
double r_value = 0.5;
double p_value = 2.0;
double q_value = 10.0;
byte object;
byte backg;
if (par1 != 0) {
IJ.log("Phansalkar: changed k_value from :" + k_value + " to:" + par1);
k_value = par1;
}
if (par2 != 0) {
IJ.log("Phansalkar: changed r_value from :" + r_value + " to:" + par2);
r_value = par2;
}
if (doIwhite) {
object = (byte) 0xff;
backg = (byte) 0;
} else {
object = (byte) 0;
backg = (byte) 0xff;
}
Meanimp = duplicateImage(ip);
ContrastEnhancer ce = new ContrastEnhancer();
ce.stretchHistogram(Meanimp, 0.0);
ImageConverter ic = new ImageConverter(Meanimp);
ic.convertToGray32();
ipMean = Meanimp.getProcessor();
ipMean.multiply(1.0 / 255);
Orimp = duplicateImage(ip);
ce.stretchHistogram(Orimp, 0.0);
ic = new ImageConverter(Orimp);
ic.convertToGray32();
ipOri = Orimp.getProcessor();
ipOri.multiply(1.0 / 255); // original to compare
// Orimp.show();
RankFilters rf = new RankFilters();
rf.rank(ipMean, radius, rf.MEAN); // Mean
// Meanimp.show();
Varimp = duplicateImage(ip);
ce.stretchHistogram(Varimp, 0.0);
ic = new ImageConverter(Varimp);
ic.convertToGray32();
ipVar = Varimp.getProcessor();
ipVar.multiply(1.0 / 255);
rf.rank(ipVar, radius, rf.VARIANCE); // Variance
ipVar.sqr(); // SD
// Varimp.show();
byte[] pixels = (byte[]) ip.getPixels();
float[] ori = (float[]) ipOri.getPixels();
float[] mean = (float[]) ipMean.getPixels();
float[] sd = (float[]) ipVar.getPixels();
for (int i = 0; i < pixels.length; i++)
pixels[i] =
((ori[i])
> (mean[i]
* (1.0
+ p_value * Math.exp(-q_value * mean[i])
+ k_value * ((sd[i] / r_value) - 1.0))))
? object
: backg;
// imp.updateAndDraw();
return;
}
void Otsu(ImagePlus imp, int radius, double par1, double par2, boolean doIwhite) {
// Otsu's threshold algorithm
// C++ code by Jordan Bevik <*****@*****.**>
// ported to ImageJ plugin by G.Landini. Same algorithm as in Auto_Threshold, this time on local
// circular regions
int[] data;
int w = imp.getWidth();
int h = imp.getHeight();
int position;
int radiusx2 = radius * 2;
ImageProcessor ip = imp.getProcessor();
byte[] pixels = (byte[]) ip.getPixels();
byte[] pixelsOut =
new byte
[pixels.length]; // need this to avoid changing the image data (and further histograms)
byte object;
byte backg;
if (doIwhite) {
object = (byte) 0xff;
backg = (byte) 0;
} else {
object = (byte) 0;
backg = (byte) 0xff;
}
int k, kStar; // k = the current threshold; kStar = optimal threshold
int N1, N; // N1 = # points with intensity <=k; N = total number of points
double BCV, BCVmax; // The current Between Class Variance and maximum BCV
double num, denom; // temporary bookeeping
int Sk; // The total intensity for all histogram points <=k
int S,
L =
256; // The total intensity of the image. Need to hange here if modifying for >8 bits
// images
int roiy;
Roi roi = new OvalRoi(0, 0, radiusx2, radiusx2);
// ip.setRoi(roi);
for (int y = 0; y < h; y++) {
IJ.showProgress(
(double) (y) / (h - 1)); // this method is slow, so let's show the progress bar
roiy = y - radius;
for (int x = 0; x < w; x++) {
roi.setLocation(x - radius, roiy);
ip.setRoi(roi);
// ip.setRoi(new OvalRoi(x-radius, roiy, radiusx2, radiusx2));
position = x + y * w;
data = ip.getHistogram();
// Initialize values:
S = N = 0;
for (k = 0; k < L; k++) {
S += k * data[k]; // Total histogram intensity
N += data[k]; // Total number of data points
}
Sk = 0;
N1 = data[0]; // The entry for zero intensity
BCV = 0;
BCVmax = 0;
kStar = 0;
// Look at each possible threshold value,
// calculate the between-class variance, and decide if it's a max
for (k = 1; k < L - 1; k++) { // No need to check endpoints k = 0 or k = L-1
Sk += k * data[k];
N1 += data[k];
// The float casting here is to avoid compiler warning about loss of precision and
// will prevent overflow in the case of large saturated images
denom = (double) (N1) * (N - N1); // Maximum value of denom is (N^2)/4 = approx. 3E10
if (denom != 0) {
// Float here is to avoid loss of precision when dividing
num = ((double) N1 / N) * S - Sk; // Maximum value of num = 255*N = approx 8E7
BCV = (num * num) / denom;
} else BCV = 0;
if (BCV >= BCVmax) { // Assign the best threshold found so far
BCVmax = BCV;
kStar = k;
}
}
// kStar += 1; // Use QTI convention that intensity -> 1 if intensity >= k
// (the algorithm was developed for I-> 1 if I <= k.)
// return kStar;
pixelsOut[position] = ((int) (pixels[position] & 0xff) > kStar) ? object : backg;
}
}
for (position = 0; position < w * h; position++)
pixels[position] = pixelsOut[position]; // update with thresholded pixels
}
void Niblack(ImagePlus imp, int radius, double par1, double par2, boolean doIwhite) {
// Niblack recommends K_VALUE = -0.2 for images with black foreground
// objects, and K_VALUE = +0.2 for images with white foreground objects.
// Niblack W. (1986) "An introduction to Digital Image Processing" Prentice-Hall.
// Ported to ImageJ plugin from E Celebi's fourier_0.8 routines
// This version uses a circular local window, instead of a rectagular one
ImagePlus Meanimp, Varimp;
ImageProcessor ip = imp.getProcessor(), ipMean, ipVar;
double k_value;
int c_value = 0;
byte object;
byte backg;
if (doIwhite) {
k_value = 0.2;
object = (byte) 0xff;
backg = (byte) 0;
} else {
k_value = -0.2;
object = (byte) 0;
backg = (byte) 0xff;
}
if (par1 != 0) {
IJ.log("Niblack: changed k_value from :" + k_value + " to:" + par1);
k_value = par1;
}
if (par2 != 0) {
IJ.log(
"Niblack: changed c_value from :"
+ c_value
+ " to:"
+ par2); // requested feature, not in original
c_value = (int) par2;
}
Meanimp = duplicateImage(ip);
ImageConverter ic = new ImageConverter(Meanimp);
ic.convertToGray32();
ipMean = Meanimp.getProcessor();
RankFilters rf = new RankFilters();
rf.rank(ipMean, radius, rf.MEAN); // Mean
// Meanimp.show();
Varimp = duplicateImage(ip);
ic = new ImageConverter(Varimp);
ic.convertToGray32();
ipVar = Varimp.getProcessor();
rf.rank(ipVar, radius, rf.VARIANCE); // Variance
// Varimp.show();
byte[] pixels = (byte[]) ip.getPixels();
float[] mean = (float[]) ipMean.getPixels();
float[] var = (float[]) ipVar.getPixels();
for (int i = 0; i < pixels.length; i++)
pixels[i] =
((int) (pixels[i] & 0xff) > (int) (mean[i] + k_value * Math.sqrt(var[i]) - c_value))
? object
: backg;
// imp.updateAndDraw();
return;
}
