def histogram(source, selem=None, sink=None, mask=None, max_bin=None):
"""Normalized sliding window histogram
Arguments
---------
source : array
Input array.
selem : array
Structure element. If None, use a cube of size 3.
sink : array
Output array. If None, a new array is allocated.
mask : array
Optional mask, if None, the complete source is used.
Pixels on the mask are zero in the output.
max_bin : int or None
Maximal number of bins.
Returns
-------
sink : array
Array of the source shape pluse on extra dimension for the histogram
at each pixel.
"""
if max_bin is None:
max_bin = io.max_value(source.dtype)
if max_bin >= 2**16:
raise ValueError(
'The histograms are to large for this code to be efficient!')
parameter_index = [max_bin]
return _apply_code(code.histogram,
code.histogram_masked,
max_bin=max_bin,
sink_shape_per_pixel=(max_bin, ),
sink_dtype=float,
source=source,
selem=selem,
sink=sink,
mask=mask,
parameter_index=parameter_index)
def _apply_code(function,
function_mask,
source,
selem=None,
sink=None,
sink_dtype=None,
sink_shape_per_pixel=None,
mask=None,
max_bin=None,
parameter_index=None,
parameter_float=None):
"""Helper to apply code."""
selem = _initialize_selem(selem, source.ndim)
if source.ndim > 3:
raise ValueError('Source dimension %d not supported!' % source.ndim)
if source.ndim != 3:
source = np.asarray(source)
shape_remove = [d for d in range(source.ndim, 3)]
source = source.view()
source.shape = source.shape + (1, ) * (3 - source.ndim)
if mask is not None:
mask = np.asarray(mask)
mask = mask.view()
mask.shape = mask.shape + (1, ) * (3 - mask.ndim)
selem = selem.view()
selem.shape = selem.shape + (1, ) * (3 - selem.ndim)
else:
shape_remove = []
if source.dtype not in (np.uint8, np.uint16, np.int, np.int16, np.int32,
np.int64, np.uint32, np.uint64, np.bool):
raise ValueError(
'The rank filter requires a source of integer type, found %r!' %
source.dtype)
if mask is not None:
if mask.dtype == bool:
mask = mask.view('uint8')
if mask.shape != source.shape:
raise ValueError(
'Source shape %r and mask shape %r do not match!' %
(source.shape, mask.shape))
if source is sink:
raise ValueError("Cannot perform rank filter in place!")
if sink_shape_per_pixel is None:
shape_per_pixel = (1, )
shape_remove += [3]
else:
shape_per_pixel = sink_shape_per_pixel
if sink is None:
if sink_dtype is None:
sink_dtype = source.dtype
sink = np.zeros(source.shape + shape_per_pixel, dtype=sink_dtype)
else:
if shape_per_pixel != (1, ):
if sink.shape != source.shape + shape_per_pixel:
raise ValueError(
'The sink of shape %r does not have expected shape %r!' %
(sink.shape, source.shape + shape_per_pixel))
if sink.shape != source.shape + shape_per_pixel:
sink = sink.reshape(source.shape + shape_per_pixel)
if sink.dtype == bool:
s = sink.view('uint8')
else:
s = sink
if max_bin is None:
max_bin = io.max_value(source.dtype)
if max_bin >= 2**16:
raise ValueError(
'The histograms are to large for this code to be efficient!')
bitdepth = int(np.log2(max_bin))
if bitdepth > 12:
warnings.warn(
"Bitdepth of %d may result in bad rank filter performance." %
bitdepth)
if parameter_index is None:
parameter_index = np.zeros(0, dtype=int)
parameter_index = np.asarray([parameter_index], dtype=int).flatten()
if parameter_float is None:
parameter_float = np.zeros(0, dtype=float)
parameter_float = np.asarray([parameter_float], dtype=float).flatten()
#print(source.__class__, source.dtype, s.__class__, s.dtype, selem.dtype, mask.dtype, max_bin)
if mask is None:
function(source=source,
selem=selem,
sink=s,
max_bin=max_bin,
p=parameter_index,
q=parameter_float)
else:
function_mask(source=source,
selem=selem,
mask=mask,
sink=s,
max_bin=max_bin,
p=parameter_index,
q=parameter_float)
if len(shape_remove) > 0:
shape = tuple(s for d, s in enumerate(sink.shape)
if d not in shape_remove)
sink = sink.reshape(shape)
return sink
