accumulator_mean_stack.py

Go to the documentation of this file.

# This file is part of meas_algorithms.
#
# LSST Data Management System
# This product includes software developed by the
# LSST Project (http://www.lsst.org/).
# See COPYRIGHT file at the top of the source tree.
#
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# You should have received a copy of the LSST License Statement and
# the GNU General Public License along with this program.  If not,
# see <https://www.lsstcorp.org/LegalNotices/>.
#
import numpy as np
 
 
__all__ = ['AccumulatorMeanStack']
 
 
class AccumulatorMeanStack(object):
    """Stack masked images.
 
    Parameters
    ----------
    shape : `tuple`
        Shape of the input and output images.
    bit_mask_value : `int`
        Bit mask to flag for "bad" inputs that should not be stacked.
    mask_threshold_dict : `dict` [`int`: `float`], optional
        Dictionary of mapping from bit number to threshold for flagging.
        Only bad bits (in bit_mask_value) which mask fractional weight
        greater than this threshold will be flagged in the output image.
    mask_map : `list` [`tuple`], optional
        Mapping from input image bits to aggregated coadd bits.
    no_good_pixels_mask : `int`, optional
        Bit mask to set when there are no good pixels in the stack.
        If not set then will set coadd masked image 'NO_DATA' bit.
    calc_error_from_input_variance : `bool`, optional
        Calculate the error from the input variance?
    compute_n_image : `bool`, optional
        Calculate the n_image map as well as stack?
    """
    def __init__(self, shape,
                 bit_mask_value, mask_threshold_dict={},
                 mask_map=[], no_good_pixels_mask=None,
                 calc_error_from_input_variance=True,
                 compute_n_image=False):
        self.shape = shape
        self.bit_mask_value = bit_mask_value
        self.mask_map = mask_map
        self.no_good_pixels_mask = no_good_pixels_mask
        self.calc_error_from_input_variance = calc_error_from_input_variance
        self.compute_n_image = compute_n_image
 
        # Only track threshold bits that are in the bad bit_mask_value.
        self.mask_threshold_dict = {}
        for bit in mask_threshold_dict:
            if (self.bit_mask_value & 2**bit) > 0:
                self.mask_threshold_dict[bit] = mask_threshold_dict[bit]
 
        # sum_weight holds the sum of weights for each pixel.
        self.sum_weight = np.zeros(shape, dtype=np.float64)
        # sum_wdata holds the sum of weight*data for each pixel.
        self.sum_wdata = np.zeros(shape, dtype=np.float64)
 
        if calc_error_from_input_variance:
            # sum_w2var holds the sum of weight**2 * variance for each pixel.
            self.sum_w2var = np.zeros(shape, dtype=np.float64)
        else:
            # sum_weight2 holds the sum of weight**2 for each pixel.
            self.sum_weight2 = np.zeros(shape, dtype=np.float64)
            # sum_wdata2 holds the sum of weight * data**2 for each pixel.
            self.sum_wdata2 = np.zeros(shape, dtype=np.float64)
 
        self.or_mask = np.zeros(shape, dtype=np.int64)
        self.rejected_weights_by_bit = {}
        for bit in self.mask_threshold_dict:
            self.rejected_weights_by_bit[bit] = np.zeros(shape, dtype=np.float64)
 
        self.masked_pixels_mask = np.zeros(shape, dtype=np.int64)
 
        if self.compute_n_image:
            self.n_image = np.zeros(shape, dtype=np.int32)
 
    def add_masked_image(self, masked_image, weight=1.0):
        """Add a masked image to the stack.
 
        Parameters
        ----------
        masked_image : `lsst.afw.image.MaskedImage`
            Masked image to add to the stack.
        weight : `float` or `np.ndarray`, optional
            Weight to apply for weighted mean.  If an array,
            must be same size and shape as input masked_image.
        """
        good_pixels = np.where(((masked_image.mask.array & self.bit_mask_value) == 0)
                               & np.isfinite(masked_image.mask.array))
 
        self.sum_weight[good_pixels] += weight
        self.sum_wdata[good_pixels] += weight*masked_image.image.array[good_pixels]
 
        if self.compute_n_image:
            self.n_image[good_pixels] += 1
 
        if self.calc_error_from_input_variance:
            self.sum_w2var[good_pixels] += (weight**2.)*masked_image.variance.array[good_pixels]
        else:
            self.sum_weight2[good_pixels] += weight**2.
            self.sum_wdata2[good_pixels] += weight*(masked_image.image.array[good_pixels]**2.)
 
        # Mask bits are propagated for good pixels
        self.or_mask[good_pixels] |= masked_image.mask.array[good_pixels]
 
        # Bad pixels are only tracked if they cross a threshold
        for bit in self.mask_threshold_dict:
            bad_pixels = ((masked_image.mask.array & 2**bit) > 0)
            self.rejected_weights_by_bit[bit][bad_pixels] += weight
            self.masked_pixels_mask[bad_pixels] |= 2**bit
 
    def fill_stacked_masked_image(self, stacked_masked_image):
        """Fill the stacked mask image after accumulation.
 
        Parameters
        ----------
        stacked_masked_image : `lsst.afw.image.MaskedImage`
            Total masked image.
        """
        with np.warnings.catch_warnings():
            # Let the NaNs through and flag bad pixels below
            np.warnings.simplefilter("ignore")
 
            # The image plane is sum(weight*data)/sum(weight)
            stacked_masked_image.image.array[:, :] = self.sum_wdata/self.sum_weight
 
            if self.calc_error_from_input_variance:
                mean_var = self.sum_w2var/(self.sum_weight**2.)
            else:
                # Compute the biased estimator
                variance = self.sum_wdata2/self.sum_weight - stacked_masked_image.image.array[:, :]**2.
                # De-bias
                variance *= (self.sum_weight**2.)/(self.sum_weight**2. - self.sum_weight2)
 
                # Compute the mean variance
                mean_var = variance*self.sum_weight2/(self.sum_weight**2.)
 
            stacked_masked_image.variance.array[:, :] = mean_var
 
            # Propagate bits when they cross the threshold
            for bit in self.mask_threshold_dict:
                hypothetical_total_weight = self.sum_weight + self.rejected_weights_by_bit[bit]
                self.rejected_weights_by_bit[bit] /= hypothetical_total_weight
                propagate = np.where(self.rejected_weights_by_bit[bit] > self.mask_threshold_dict[bit])
                self.or_mask[propagate] |= 2**bit
 
            # Map mask planes to new bits for pixels that had at least one
            # bad input rejected and are in the mask_map.
            for mask_tuple in self.mask_map:
                self.or_mask[(self.masked_pixels_mask & mask_tuple[0]) > 0] |= mask_tuple[1]
 
            stacked_masked_image.mask.array[:, :] = self.or_mask
 
        if self.no_good_pixels_mask is None:
            mask_dict = stacked_masked_image.mask.getMaskPlaneDict()
            no_good_pixels_mask = 2**(mask_dict['NO_DATA'])
        else:
            no_good_pixels_mask = self.no_good_pixels_mask
 
        bad_pixels = (self.sum_weight <= 0.0)
        stacked_masked_image.mask.array[bad_pixels] |= no_good_pixels_mask
 
    def add_image(self, image, weight=1.0):
        """Add an image to the stack.
 
        No bit-filtering is performed when adding an image.
 
        Parameters
        ----------
        image : `lsst.afw.image.Image`
            Image to add to the stack.
        weight : `float` or `np.ndarray`, optional
            Weight to apply for weighted mean.  If an array,
            must be same size and shape as input image.
        """
        self.sum_weight[:, :] += weight
        self.sum_wdata[:, :] += weight*image.array[:]
 
        if self.compute_n_image:
            self.n_image[:, :] += 1
 
    def fill_stacked_image(self, stacked_image):
        """Fill the image after accumulation.
 
        Parameters
        ----------
        stacked_image : `lsst.afw.image.Image`
            Total image.
        """
        with np.warnings.catch_warnings():
            # Let the NaNs through, this should only happen
            # if we're stacking with no inputs.
            np.warnings.simplefilter("ignore")
 
            # The image plane is sum(weight*data)/sum(weight)
            stacked_image.array[:, :] = self.sum_wdata/self.sum_weight
 
    @staticmethod
    def stats_ctrl_to_threshold_dict(stats_ctrl):
        """Convert stats control to threshold dict.
 
        Parameters
        ----------
        stats_ctrl : `lsst.afw.math.StatisticsControl`
 
        Returns
        -------
        threshold_dict : `dict`
            Dict mapping from bit to propagation threshold.
        """
        threshold_dict = {}
        for bit in range(64):
            threshold_dict[bit] = stats_ctrl.getMaskPropagationThreshold(bit)
 
        return threshold_dict