Implement find_objects for Dask Array

Provides an implementation of `find_objects` that determines bounding
boxes for each label in the image. Currently requires the user to
specify the number of labels they would like to inspect. Raises a
`NotImplementedError` if the user wishes to collect all bounding boxes
for labels. Works by selecting the 1-D positions that correspond to the
label while ignoring all other points. Assumes that these positions
along with an intermediate array of the same size comfortably fit in
memory.

Within a utility function, determines whether any positions were found
for the corresponding label. If not, simply returns `None`. If positions
were found, it manually unravels the positions and finds the maximum and
minimum positions along each dimension. These are stored into `slice`s,
which are packed into a `tuple` and returned. Makes sure to use in-place
NumPy operations to avoid using additional memory.

dask_image/ndmeasure/__init__.py

@@ -127,6 +127,46 @@ def extrema(input, labels=None, index=None):
     return result
 
 
+def find_objects(input, max_label=0):
+    """
+    Find the center of mass over an image at specified subregions.
+
+    Parameters
+    ----------
+    input : ndarray
+        Image features noted by integers.
+    max_label : int, optional
+        Maximum label to look for in ``input``. If 0, look for all labels.
+
+    Returns
+    -------
+    object_slices : ``list`` of ``tuple``s
+        A ``list`` of ``tuple``s specifying the bounding boxes of each label.
+    """
+
+    # Normalize arguments
+    input = dask.array.asarray(input)
+    max_label = int(max_label)
+
+    # Catch unsupported case
+    if max_label == 0:
+        raise NotImplementedError(
+            "Getting all labels is currently not supported."
+        )
+
+    positions = _utils._ravel_shape_indices(
+        input.shape, chunks=input.chunks
+    )
+
+    object_slices = []
+    for i in range(1, max_label + 1):
+        object_slices.append(_utils._find_object(
+            positions[input == i], input.shape
+        ))
+
+    return object_slices
+
+
 def histogram(input,
               min,
               max,

dask_image/ndmeasure/_utils.py

@@ -141,6 +141,28 @@ def _extrema(a, positions, shape, dtype):
     return result[0]
 
 
+@dask.delayed
+def _find_object(pos, shape):
+    """
+    Find bounding box for the given positions.
+    """
+
+    if pos.size == 0:
+        return None
+
+    pos = numpy.require(pos, requirements='OW')
+
+    result = []
+    pos_i = numpy.empty_like(pos)
+    for s in reversed(shape):
+        numpy.mod(pos, s, out=pos_i)
+        numpy.subtract(pos, pos_i, out=pos)
+        result.insert(0, slice(pos_i.min(), pos_i.max() + 1))
+    result = tuple(result)
+
+    return result
+
+
 def _histogram(input,
                min,
                max,