Implementing "ImageClass" (python) class in PlantCV

[DannieSheng]

We've discussed implementing "ImageClass" in PlantCV by subclassing ndarray (attaching attributes to images as python NumPy arrays).
Some potential attributes are:

• filename
• imgType (acceptable sample values: rgb, bgr)
• (optional) mask
• (optional) image of another datatype for specific purposes (similar to the pseuco_RGB image attached to a hyperspectral image)

The motivation/goal for implementing "ImageClass" :

1. Having specific attributes attached to images themselves such that:
   1)Some functions can take fewer input variables
   2)Some checks (e.g. datatype checks, rgb or bgr checks) can be done internally
2. Easier for developers, potentially won't break a lot of the usage for users

Everybody is welcomed to brainstorm and add attributes. And please feel free to give suggestions to this implementation by thinking about questions like

• whether this implementation over-complicates things
• other goals to be achieved from this implementation

[nfahlgren]

This is the prototype class I put together previously that just adds imgtype and filename attributes. It also automatically changes the imgtype to GRAY when the dimensions decrease from 3 to 2:

class Image (np.ndarray):
	# The Image class extends the np.ndarray class by adding attributes.
	# Documentation from NumPy here: 
	def __new__(cls, input_array, imgtype, filename):
		obj = np.asarray(input_array).view(cls)
		# New attribute imgtype stores the image type metdata
		# (e.g. BGR, RGB, GRAY, etc.)
		obj.imgtype = imgtype
		# New attribute filename stores the path and filename 
		# of the source file
		obj.filename = filename
		return obj

	def __array_finalize__(self, obj):
		if obj is None: return
		self.imgtype = getattr(obj, "imgtype", None)
		self.filename = getattr(obj, "filename", None)

	def __getitem__(self, key):
		# Enhance the np.ndarray __getitem__ method
		# Slice the array as requested but return an array of the same class
		# Idea from NumPy examples of subclassing: 
		value = super(Image, self).__getitem__(key)
		# If the slice reduces the dimensions of the ndarray from 3 to 2, 
		# set imgtype to GRAY
		# Note that we can also handle this when we apply functions to 
		# an image, but need to handle user manipulation too
		if len(self.shape) == 3 and len(value.shape) == 2:
			value.imgtype = "GRAY"
		return value