Downsample a 2D image array

downsample.py

def downsample(data, downsampling, summary=np.sum, allow_trim=False):
    """Downsample a 2D array.
    
    Parameters
    ----------
    data : array
        Two dimensional array of values to downsample.
    downsampling : int
        Downsampling factor to use along both dimensions. Must evenly divide the
        data dimensions when allow_trim is False.
    summary : callable
        The summary function to use that will be applied to each block of shape
        (dowsampling, downsampling) to obtain the output downsampled values.
        Must support broadcasting and an axis parameter. Useful choices are
        np.sum, np.mean, np.min, np.max, np.median, np.var but any ufunc
        should work.
    allow_trim : bool
        When False, the input dimensions (ny, nx) must both exactly divide
        the downsampling value.  Otherwise, any extra rows and columns are
        silently trimmed before apply the summary function.
        
    Returns
    -------
    array
        A two dimensional array of shape (ny // downsampling, nx // downsampling)
        where the input data shape is (ny, nx).
    """
    data = np.asarray(data)
    if data.ndim != 2:
        raise ValueError('Data must be 2 dimensional.')
    ny, nx = data.shape
    if not allow_trim and ((nx % downsampling) or (ny % downsampling)):
        raise ValueError('Data shape {0} does not evenly divide downsampling={1} and allow_trim is False.'
                         .format((ny, nx), downsampling))
    ny //= downsampling
    nx //= downsampling
    shape = (ny, nx, downsampling, downsampling)
    strides = (downsampling * data.strides[0], downsampling * data.strides[1]) + data.strides
    blocks = np.lib.stride_tricks.as_strided(
        data[:downsampling * ny, :downsampling * nx], shape=shape, strides=strides)
    return summary(blocks, axis=(2, 3))