Mercurial > repos > imgteam > imagej2_find_maxima
view imagej2_find_maxima_jython_script.py @ 3:4409f9eb9433 draft default tip
planemo upload for repository https://github.com/bgruening/galaxytools/tree/master/tools/image_processing/imagej2 commit 8f49f3c66b5a1de99ec15e65c2519a56792f1d56
author | imgteam |
---|---|
date | Wed, 25 Sep 2024 15:58:30 +0000 |
parents | 6a07f71806bb |
children |
line wrap: on
line source
import sys from ij import IJ, ImagePlus from ij.plugin.filter import Analyzer, MaximumFinder from ij.process import ImageProcessor # Fiji Jython interpreter implements Python 2.5 which does not # provide support for argparse. input_file = sys.argv[-9] scale_when_converting = sys.argv[-8] == "yes" weighted_rgb_conversions = sys.argv[-7] == "yes" noise_tolerance = int(sys.argv[-6]) output_type = sys.argv[-5] exclude_edge_maxima = sys.argv[-4] == "yes" light_background = sys.argv[-3] tmp_output_path = sys.argv[-2] output_datatype = sys.argv[-1] # Open the input image file. input_image_plus = IJ.openImage(input_file) # Create a copy of the image. input_image_plus_copy = input_image_plus.duplicate() image_processor_copy = input_image_plus_copy.getProcessor() bit_depth = image_processor_copy.getBitDepth() analyzer = Analyzer(input_image_plus_copy) # Set the conversion options. options = [] # The following 2 options are applicable only to RGB images. if bit_depth == 24: if scale_when_converting: options.append("scale") if weighted_rgb_conversions: options.append("weighted") # Perform conversion - must happen even if no options are set. IJ.run(input_image_plus_copy, "Conversions...", " %s" % " ".join(options)) if output_type in ["List", "Count"]: # W're generating a tabular file for the output. # Set the Find Maxima options. options = ["noise=%d" % noise_tolerance] if output_type.find("_") > 0: output_type_str = "output=[%s]" % output_type.replace("_", " ") else: output_type_str = "output=%s" % output_type options.append(output_type_str) if exclude_edge_maxima: options.append("exclude") if light_background: options.append("light") # Run the command. IJ.run(input_image_plus_copy, "Find Maxima...", "%s" % " ".join(options)) results_table = analyzer.getResultsTable() results_table.saveAs(tmp_output_path) else: # Find the maxima of an image (does not find minima). # LIMITATIONS: With output_type=Segmented_Particles # (watershed segmentation), some segmentation lines # may be improperly placed if local maxima are suppressed # by the tolerance. mf = MaximumFinder() if output_type == "Single_Points": output_type_param = mf.SINGLE_POINTS elif output_type == "Maxima_Within_Tolerance": output_type_param = mf.IN_TOLERANCE elif output_type == "Segmented_Particles": output_type_param = mf.SEGMENTED elif output_type == "List": output_type_param = mf.LIST elif output_type == "Count": output_type_param = mf.COUNT # Get a new byteProcessor with a normal (uninverted) LUT where # the marked points are set to 255 (Background 0). Pixels outside # of the roi of the input image_processor_copy are not set. No # output image is created for output types POINT_SELECTION, LIST # and COUNT. In these cases findMaxima returns null. byte_processor = mf.findMaxima( image_processor_copy, noise_tolerance, ImageProcessor.NO_THRESHOLD, output_type_param, exclude_edge_maxima, False, ) # Invert the image or ROI. byte_processor.invert() if output_type == "Segmented_Particles" and not light_background: # Invert the values in this image's LUT (indexed color model). byte_processor.invertLut() image_plus = ImagePlus("output", byte_processor) IJ.saveAs(image_plus, output_datatype, tmp_output_path)