Geografic-Information-System/scripts/i.pansharpen/i.pansharpen.py

#!/usr/bin/env python3

############################################################################
#
# MODULE:	    i.pansharpen
#
# AUTHOR(S):    Overall script by Michael Barton (ASU)
#               Brovey transformation in i.fusion.brovey by Markus Neteler <<neteler at osgeo org>>
#               i.fusion brovey converted to Python by Glynn Clements
#               IHS and PCA transformation added by Michael Barton (ASU)
#               histogram matching algorithm by Michael Barton and Luca Delucchi, Fondazione E. Mach (Italy)
#               Thanks to Markus Metz for help with PCA inversion
#               Thanks to Hamish Bowman for parallel processing algorithm
#
# PURPOSE:	Sharpening of 3 RGB channels using a high-resolution panchromatic channel
#
# COPYRIGHT:	(C) 2002-2019 by the GRASS Development Team
#
#		This program is free software under the GNU General Public
#		License (>=v2). Read the file COPYING that comes with GRASS
#		for details.
#
# REFERENCES:
#   Roller, N.E.G. and Cox, S., 1980. Comparison of Landsat MSS and merged MSS/RBV
#   data for analysis of natural vegetation. Proc. of the 14th International
#   Symposium on Remote Sensing of Environment, San Jose, Costa Rica, 23-30 April, pp. 1001-1007.
#
#   Amarsaikhan, D., & Douglas, T. (2004). Data fusion and multisource image classification.
#   International Journal of Remote Sensing, 25(17), 3529-3539.
#
#   Behnia, P. (2005). Comparison between four methods for data fusion of ETM+
#   multispectral and pan images. Geo-spatial Information Science, 8(2), 98-103
#
#   for LANDSAT 5: see Pohl, C 1996 and others
#
#############################################################################

#%Module
#% description: Image fusion algorithms to sharpen multispectral with high-res panchromatic channels
#% keyword: imagery
#% keyword: fusion
#% keyword: sharpen
#% keyword: Brovey
#% keyword: IHS
#% keyword: HIS
#% keyword: PCA
#% overwrite: yes
#%End
#%option G_OPT_R_INPUT
#% key: red
#% description: Name of raster map to be used for <red>
#%end
#%option G_OPT_R_INPUT
#% key: green
#% description: Name of raster map to be used for <green>
#%end
#%option G_OPT_R_INPUT
#% key: blue
#% description: Name of raster map to be used for <blue>
#%end
#% option G_OPT_R_INPUT
#% key: pan
#% description: Name of raster map to be used for high resolution panchromatic channel
#%end
#%option G_OPT_R_BASENAME_OUTPUT
#%end
#%option
#% key: method
#% description: Method for pan sharpening
#% options: brovey,ihs,pca
#% answer: ihs
#% required: yes
#%end
#%option
#% key: bitdepth
#% type: integer
#% description: Bit depth of image (must be in range of 2-30)
#% options: 2-32
#% answer: 8
#% required: yes
#%end
#%flag
#% key: s
#% description: Serial processing rather than parallel processing
#%end
#%flag
#% key: l
#% description: Rebalance blue channel for LANDSAT
#%end
#%flag
#% key: r
#% description: Rescale (stretch) the range of pixel values in each channel to the entire 0-255 8-bit range for processing (see notes)
#%end

import os

try:
    import numpy as np
    hasNumPy = True
except ImportError:
    hasNumPy = False

import grass.script as grass


def main():
    if not hasNumPy:
        grass.fatal(_("Required dependency NumPy not found. Exiting."))

    sharpen   = options['method']  # sharpening algorithm
    ms1_orig  = options['blue']  # blue channel
    ms2_orig  = options['green']  # green channel
    ms3_orig  = options['red']  # red channel
    pan_orig  = options['pan']  # high res pan channel
    out       = options['output']  # prefix for output RGB maps
    bits      = options['bitdepth'] # bit depth of image channels
    bladjust  = flags['l']  # adjust blue channel
    sproc     = flags['s']  # serial processing
    rescale   = flags['r'] # rescale to spread pixel values to entire 0-255 range

    # Checking bit depth
    bits = float(bits)
    if bits < 2 or bits > 30:
        grass.warning(_("Bit depth is outside acceptable range"))
        return

    outb = grass.core.find_file('%s_blue' % out)
    outg = grass.core.find_file('%s_green' % out)
    outr = grass.core.find_file('%s_red' % out)

    if (outb['name'] != '' or outg['name'] != '' or outr['name'] != '') and not grass.overwrite():
        grass.warning(_('Maps with selected output prefix names already exist.'
                        ' Delete them or use overwrite flag'))
        return

    pid = str(os.getpid())

    # convert input image channels to 8 bit for processing
    ms1 = 'tmp%s_ms1' % pid
    ms2 = 'tmp%s_ms2' % pid
    ms3 = 'tmp%s_ms3' % pid
    pan = 'tmp%s_pan' % pid

    if rescale == False:
        if bits == 8:
            grass.message(_("Using 8bit image channels"))
            if sproc:
                # serial processing
                grass.run_command('g.copy', raster='%s,%s' % (ms1_orig, ms1),
                                   quiet=True, overwrite=True)
                grass.run_command('g.copy', raster='%s,%s' % (ms2_orig, ms2),
                                   quiet=True, overwrite=True)
                grass.run_command('g.copy', raster='%s,%s' % (ms3_orig, ms3),
                                   quiet=True, overwrite=True)
                grass.run_command('g.copy', raster='%s,%s' % (pan_orig, pan),
                                   quiet=True, overwrite=True)
            else:
                # parallel processing
                pb = grass.start_command('g.copy', raster='%s,%s' % (ms1_orig, ms1),
                                       quiet=True, overwrite=True)
                pg = grass.start_command('g.copy', raster='%s,%s' % (ms2_orig, ms2),
                                       quiet=True, overwrite=True)
                pr = grass.start_command('g.copy', raster='%s,%s' % (ms3_orig, ms3),
                                       quiet=True, overwrite=True)
                pp = grass.start_command('g.copy', raster='%s,%s' % (pan_orig, pan),
                                       quiet=True, overwrite=True)

                pb.wait()
                pg.wait()
                pr.wait()
                pp.wait()

        else:
            grass.message(_("Converting image chanels to 8bit for processing"))
            maxval = pow(2, bits) - 1
            if sproc:
                # serial processing
                grass.run_command('r.rescale', input=ms1_orig, from_='0,%f' % maxval,
                                   output=ms1, to='0,255', quiet=True, overwrite=True)
                grass.run_command('r.rescale', input=ms2_orig, from_='0,%f' % maxval,
                                   output=ms2, to='0,255', quiet=True, overwrite=True)
                grass.run_command('r.rescale', input=ms3_orig, from_='0,%f' % maxval,
                                   output=ms3, to='0,255', quiet=True, overwrite=True)
                grass.run_command('r.rescale', input=pan_orig, from_='0,%f' % maxval,
                                   output=pan, to='0,255', quiet=True, overwrite=True)

            else:
                # parallel processing
                pb = grass.start_command('r.rescale', input=ms1_orig, from_='0,%f' % maxval,
                                       output=ms1, to='0,255', quiet=True, overwrite=True)
                pg = grass.start_command('r.rescale', input=ms2_orig, from_='0,%f' % maxval,
                                       output=ms2, to='0,255', quiet=True, overwrite=True)
                pr = grass.start_command('r.rescale', input=ms3_orig, from_='0,%f' % maxval,
                                       output=ms3, to='0,255', quiet=True, overwrite=True)
                pp = grass.start_command('r.rescale', input=pan_orig, from_='0,%f' % maxval,
                                       output=pan, to='0,255', quiet=True, overwrite=True)

                pb.wait()
                pg.wait()
                pr.wait()
                pp.wait()

    else:
        grass.message(_("Rescaling image chanels to 8bit for processing"))

        min_ms1 = int(grass.raster_info(ms1_orig)['min'])
        max_ms1 = int(grass.raster_info(ms1_orig)['max'])
        min_ms2 = int(grass.raster_info(ms2_orig)['min'])
        max_ms2 = int(grass.raster_info(ms2_orig)['max'])
        min_ms3 = int(grass.raster_info(ms3_orig)['min'])
        max_ms3 = int(grass.raster_info(ms3_orig)['max'])
        min_pan = int(grass.raster_info(pan_orig)['min'])
        max_pan = int(grass.raster_info(pan_orig)['max'])

        maxval = pow(2, bits) - 1
        if sproc:
            # serial processing
            grass.run_command('r.rescale', input=ms1_orig, from_='%f,%f' % (min_ms1, max_ms1),
                               output=ms1, to='0,255', quiet=True, overwrite=True)
            grass.run_command('r.rescale', input=ms2_orig, from_='%f,%f' % (min_ms2, max_ms2),
                               output=ms2, to='0,255', quiet=True, overwrite=True)
            grass.run_command('r.rescale', input=ms3_orig, from_='%f,%f' % (min_ms3, max_ms3),
                               output=ms3, to='0,255', quiet=True, overwrite=True)
            grass.run_command('r.rescale', input=pan_orig, from_='%f,%f' % (min_pan, max_pan),
                               output=pan, to='0,255', quiet=True, overwrite=True)

        else:
            # parallel processing
            pb = grass.start_command('r.rescale', input=ms1_orig, from_='%f,%f' % (min_ms1, max_ms1),
                                   output=ms1, to='0,255', quiet=True, overwrite=True)
            pg = grass.start_command('r.rescale', input=ms2_orig, from_='%f,%f' % (min_ms2, max_ms2),
                                   output=ms2, to='0,255', quiet=True, overwrite=True)
            pr = grass.start_command('r.rescale', input=ms3_orig, from_='%f,%f' % (min_ms3, max_ms3),
                                   output=ms3, to='0,255', quiet=True, overwrite=True)
            pp = grass.start_command('r.rescale', input=pan_orig, from_='%f,%f' % (min_pan, max_pan),
                                   output=pan, to='0,255', quiet=True, overwrite=True)

            pb.wait()
            pg.wait()
            pr.wait()
            pp.wait()


    # get PAN resolution:
    kv = grass.raster_info(map=pan)
    nsres = kv['nsres']
    ewres = kv['ewres']
    panres = (nsres + ewres) / 2

    # clone current region
    grass.use_temp_region()
    grass.run_command('g.region', res=panres, align=pan)

    # Select sharpening method
    grass.message(_("Performing pan sharpening with hi res pan image: %f" % panres))
    if sharpen == "brovey":
        brovey(pan, ms1, ms2, ms3, out, pid, sproc)
    elif sharpen == "ihs":
        ihs(pan, ms1, ms2, ms3, out, pid, sproc)
    elif sharpen == "pca":
        pca(pan, ms1, ms2, ms3, out, pid, sproc)
    # Could add other sharpening algorithms here, e.g. wavelet transformation

    grass.message(_("Assigning grey equalized color tables to output images..."))

    # equalized grey scales give best contrast
    grass.message(_("setting pan-sharpened channels to equalized grey scale"))
    for ch in ['red', 'green', 'blue']:
        grass.run_command('r.colors', quiet=True, map="%s_%s" % (out, ch),
                          flags="e", color='grey')

    # Landsat too blue-ish because panchromatic band less sensitive to blue
    # light, so output blue channed can be modified
    if bladjust:
        grass.message(_("Adjusting blue channel color table..."))
        blue_colors = ['0 0 0 0\n5% 0 0 0\n67% 255 255 255\n100% 255 255 255']
        # these previous colors are way too blue for landsat
        # blue_colors = ['0 0 0 0\n10% 0 0 0\n20% 200 200 200\n40% 230 230 230\n67% 255 255 255\n100% 255 255 255']
        bc = grass.feed_command('r.colors', quiet = True, map = "%s_blue" % out, rules = "-")
        bc.stdin.write(grass.encode('\n'.join(blue_colors)))
        bc.stdin.close()

    # output notice
    grass.verbose(_("The following pan-sharpened output maps have been generated:"))
    for ch in ['red', 'green', 'blue']:
        grass.verbose(_("%s_%s") % (out, ch))

    grass.verbose(_("To visualize output, run: g.region -p raster=%s_red" % out))
    grass.verbose(_("d.rgb r=%s_red g=%s_green b=%s_blue" % (out, out, out)))
    grass.verbose(_("If desired, combine channels into a single RGB map with 'r.composite'."))
    grass.verbose(_("Channel colors can be rebalanced using i.colors.enhance."))

    # write cmd history:
    for ch in ['red', 'green', 'blue']:
        grass.raster_history("%s_%s" % (out, ch))

    # create a group with the three outputs
    #grass.run_command('i.group', group=out,
    #                  input="{n}_red,{n}_blue,{n}_green".format(n=out))

    # Cleanup
    grass.message(_("cleaning up temp files"))
    try:
        grass.run_command('g.remove', flags="f", type="raster",
                           pattern="tmp%s*" % pid, quiet=True)
    except:
        ""

def brovey(pan, ms1, ms2, ms3, out, pid, sproc):
    grass.verbose(_("Using Brovey algorithm"))

    # pan/intensity histogram matching using linear regression
    grass.message(_("Pan channel/intensity histogram matching using linear regression"))
    outname = 'tmp%s_pan1' % pid
    panmatch1 = matchhist(pan, ms1, outname)

    outname = 'tmp%s_pan2' % pid
    panmatch2 = matchhist(pan, ms2, outname)

    outname = 'tmp%s_pan3' % pid
    panmatch3 = matchhist(pan, ms3, outname)

    outr = '%s_red' % out
    outg = '%s_green' % out
    outb = '%s_blue' % out

    # calculate brovey transformation
    grass.message(_("Calculating Brovey transformation..."))

    if sproc:
        # serial processing
        e = '''eval(k = "$ms1" + "$ms2" + "$ms3")
            "$outr" = 1 * round("$ms3" * "$panmatch3" / k)
            "$outg" = 1 * round("$ms2" * "$panmatch2" / k)
            "$outb" = 1 * round("$ms1" * "$panmatch1" / k)'''
        grass.mapcalc(e, outr=outr, outg=outg, outb=outb,
                      panmatch1=panmatch1, panmatch2=panmatch2,
                      panmatch3=panmatch3, ms1=ms1, ms2=ms2, ms3=ms3,
                      overwrite=True)
    else:
        # parallel processing
        pb = grass.mapcalc_start('%s_blue = 1 * round((%s * %s) / (%s + %s + %s))' %
                                 (out, ms1, panmatch1, ms1, ms2, ms3),
                                 overwrite=True)
        pg = grass.mapcalc_start('%s_green = 1 * round((%s * %s) / (%s + %s + %s))' %
                                 (out, ms2, panmatch2, ms1, ms2, ms3),
                                 overwrite=True)
        pr = grass.mapcalc_start('%s_red = 1 * round((%s * %s) / (%s + %s + %s))' %
                                 (out, ms3, panmatch3, ms1, ms2, ms3),
                                 overwrite=True)

        pb.wait(), pg.wait(), pr.wait()
        try:
            pb.terminate(), pg.terminate(), pr.terminate()
        except:
            ""

    # Cleanup
    try:
        grass.run_command('g.remove', flags='f', quiet=True, type='raster',
                          name='%s,%s,%s' % (panmatch1, panmatch2, panmatch3))
    except:
        ""

def ihs(pan, ms1, ms2, ms3, out, pid, sproc):
    grass.verbose(_("Using IHS<->RGB algorithm"))
    # transform RGB channels into IHS color space
    grass.message(_("Transforming to IHS color space..."))
    grass.run_command('i.rgb.his', overwrite=True,
                      red=ms3,
                      green=ms2,
                      blue=ms1,
                      hue="tmp%s_hue" % pid,
                      intensity="tmp%s_int" % pid,
                      saturation="tmp%s_sat" % pid)

    # pan/intensity histogram matching using linear regression
    target = "tmp%s_int" % pid
    outname = "tmp%s_pan_int" % pid
    panmatch = matchhist(pan, target, outname)

    # substitute pan for intensity channel and transform back to RGB color space
    grass.message(_("Transforming back to RGB color space and sharpening..."))
    grass.run_command('i.his.rgb', overwrite=True,
                      hue="tmp%s_hue" % pid,
                      intensity="%s" % panmatch,
                      saturation="tmp%s_sat" % pid,
                      red="%s_red" % out,
                      green="%s_green" % out,
                      blue="%s_blue" % out)

    # Cleanup
    try:
        grass.run_command('g.remove', flags='f', quiet=True, type='raster',
                          name=panmatch)
    except:
        ""

def pca(pan, ms1, ms2, ms3, out, pid, sproc):

    grass.verbose(_("Using PCA/inverse PCA algorithm"))
    grass.message(_("Creating PCA images and calculating eigenvectors..."))

    # initial PCA with RGB channels
    pca_out = grass.read_command('i.pca', quiet=True, rescale='0,0',
                                 input='%s,%s,%s' % (ms1, ms2, ms3),
                                 output='tmp%s.pca' % pid)
    if len(pca_out) < 1:
        grass.fatal(_("Input has no data. Check region settings."))

    b1evect = []
    b2evect = []
    b3evect = []
    for l in pca_out.replace('(', ',').replace(')', ',').splitlines():
        b1evect.append(float(l.split(',')[1]))
        b2evect.append(float(l.split(',')[2]))
        b3evect.append(float(l.split(',')[3]))

    # inverse PCA with hi res pan channel substituted for principal component 1
    pca1 = 'tmp%s.pca.1' % pid
    pca2 = 'tmp%s.pca.2' % pid
    pca3 = 'tmp%s.pca.3' % pid
    b1evect1 = b1evect[0]
    b1evect2 = b1evect[1]
    b1evect3 = b1evect[2]
    b2evect1 = b2evect[0]
    b2evect2 = b2evect[1]
    b2evect3 = b2evect[2]
    b3evect1 = b3evect[0]
    b3evect2 = b3evect[1]
    b3evect3 = b3evect[2]

    # Histogram matching
    outname = 'tmp%s_pan1' % pid
    panmatch1 = matchhist(pan, ms1, outname)

    outname = 'tmp%s_pan2' % pid
    panmatch2 = matchhist(pan, ms2, outname)

    outname = 'tmp%s_pan3' % pid
    panmatch3 = matchhist(pan, ms3, outname)

    grass.message(_("Performing inverse PCA ..."))

    # Get mean value of each channel
    stats1 = grass.parse_command("r.univar", map=ms1, flags='g',
                                 parse=(grass.parse_key_val,
                                        {'sep': '='}))
    stats2 = grass.parse_command("r.univar", map=ms2, flags='g',
                                 parse=(grass.parse_key_val,
                                        {'sep': '='}))
    stats3 = grass.parse_command("r.univar", map=ms3, flags='g',
                                 parse=(grass.parse_key_val,
                                        {'sep': '='}))

    b1mean = float(stats1['mean'])
    b2mean = float(stats2['mean'])
    b3mean = float(stats3['mean'])

    if sproc:
        # serial processing
        outr = '%s_red' % out
        outg = '%s_green' % out
        outb = '%s_blue' % out

        cmd1 = "$outb = 1 * round(($panmatch1 * $b1evect1) + ($pca2 * $b1evect2) + ($pca3 * $b1evect3) + $b1mean)"
        cmd2 = "$outg = 1 * round(($panmatch2 * $b2evect1) + ($pca2 * $b2evect2) + ($pca3 * $b2evect3) + $b2mean)"
        cmd3 = "$outr = 1 * round(($panmatch3 * $b3evect1) + ($pca2 * $b3evect2) + ($pca3 * $b3evect3) + $b3mean)"

        cmd = '\n'.join([cmd1, cmd2, cmd3])

        grass.mapcalc(cmd, outb=outb, outg=outg, outr=outr,
                      panmatch1=panmatch1,
                      panmatch2=panmatch2,
                      panmatch3=panmatch3,
                      pca2=pca2,
                      pca3=pca3,
                      b1evect1=b1evect1,
                      b2evect1=b2evect1,
                      b3evect1=b3evect1,
                      b1evect2=b1evect2,
                      b2evect2=b2evect2,
                      b3evect2=b3evect2,
                      b1evect3=b1evect3,
                      b2evect3=b2evect3,
                      b3evect3=b3evect3,
                      b1mean=b1mean,
                      b2mean=b2mean,
                      b3mean=b3mean,
                      overwrite=True)
    else:
        # parallel processing
        pb = grass.mapcalc_start('%s_blue = 1 * round((%s * %f) + (%s * %f) + (%s * %f) + %f)'
                                 % (out,
                                    panmatch1,
                                    b1evect1,
                                    pca2,
                                    b1evect2,
                                    pca3,
                                    b1evect3,
                                    b1mean),
                                 overwrite=True)

        pg = grass.mapcalc_start('%s_green = 1 * round((%s * %f) + (%s * %f) + (%s * %f) + %f)'
                                 % (out,
                                    panmatch2,
                                    b2evect1,
                                    pca2,
                                    b2evect2,
                                    pca3,
                                    b2evect3,
                                    b2mean),
                                 overwrite=True)

        pr = grass.mapcalc_start('%s_red = 1 * round((%s * %f) + (%s * %f) + (%s * %f) + %f)'
                                 % (out,
                                    panmatch3,
                                    b3evect1,
                                    pca2,
                                    b3evect2,
                                    pca3,
                                    b3evect3,
                                    b3mean),
                                 overwrite=True)

        pb.wait(), pg.wait(), pr.wait()
        try:
            pb.terminate(), pg.terminate(), pr.terminate()
        except:
            ""

    # Cleanup
    grass.run_command('g.remove', flags='f', quiet=True, type='raster',
                      name='%s,%s,%s' % (panmatch1, panmatch2, panmatch3))

def matchhist(original, target, matched):
    # pan/intensity histogram matching using numpy arrays
    grass.message(_("Histogram matching..."))

    # input images
    original = original.split('@')[0]
    target = target.split('@')[0]
    images = [original, target]

    # create a dictionary to hold arrays for each image
    arrays = {}

    for img in images:
        # calculate number of cells for each grey value for for each image
        stats_out = grass.pipe_command('r.stats', flags='cin', input=img,
                                       sep=':')
        stats = grass.decode(stats_out.communicate()[0]).split('\n')[:-1]
        stats_dict = dict(s.split(':', 1) for s in stats)
        total_cells = 0  # total non-null cells
        for j in stats_dict:
            stats_dict[j] = int(stats_dict[j])
            if j != '*':
                total_cells += stats_dict[j]

        if total_cells < 1:
            grass.fatal(_("Input has no data. Check region settings."))

        # Make a 2x256 structured array for each image with a
        #   cumulative distribution function (CDF) for each grey value.
        #   Grey value is the integer (i4) and cdf is float (f4).

        arrays[img] = np.zeros((256, ), dtype=('i4,f4'))
        cum_cells = 0  # cumulative total of cells for sum of current and all lower grey values

        for n in range(0, 256):
            if str(n) in stats_dict:
                num_cells = stats_dict[str(n)]
            else:
                num_cells = 0

            cum_cells += num_cells

            # cdf is the the number of cells at or below a given grey value
            #   divided by the total number of cells
            cdf = float(cum_cells) / float(total_cells)

            # insert values into array
            arrays[img][n] = (n, cdf)

    # open file for reclass rules
    outfile = open(grass.tempfile(), 'w')

    for i in arrays[original]:
        # for each grey value and corresponding cdf value in original, find the
        #   cdf value in target that is closest to the target cdf value
        difference_list = []
        for j in arrays[target]:
            # make a list of the difference between each original cdf value and
            #   the target cdf value
            difference_list.append(abs(i[1] - j[1]))

        # get the smallest difference in the list
        min_difference = min(difference_list)

        for j in arrays[target]:
            # find the grey value in target that corresponds to the cdf
            #   closest to the original cdf
            if j[1] <= i[1] + min_difference and j[1] >= i[1] - min_difference:
                # build a reclass rules file from the original grey value and
                #   corresponding grey value from target
                out_line = "%d = %d\n" % (i[0], j[0])
                outfile.write(out_line)
                break

    outfile.close()

    # create reclass of target from reclass rules file
    result = grass.core.find_file(matched, element='cell')
    if result['fullname']:
        grass.run_command('g.remove', flags='f', quiet=True, type='raster',
                          name=matched)
        grass.run_command('r.reclass', input=original, out=matched,
                          rules=outfile.name)
    else:
        grass.run_command('r.reclass', input=original, out=matched,
                          rules=outfile.name)

    # Cleanup
    # remove the rules file
    grass.try_remove(outfile.name)

    # return reclass of target with histogram that matches original
    return matched

if __name__ == "__main__":
    options, flags = grass.parser()
    main()