#!/usr/bin/env python #---------------------------------------------------------------------------- # Name: BBox.py # Purpose: # # Author: # # Created: # Version: # Date: # Licence: # Tags: phoenix-port #---------------------------------------------------------------------------- """ A Bounding Box object and assorted utilities , subclassed from a numpy array """ import numpy as N class BBox(N.ndarray): """ A Bounding Box object: Takes Data as an array. Data is any python sequence that can be turned into a 2x2 numpy array of floats:: [ [MinX, MinY ], [MaxX, MaxY ] ] It is a subclass of numpy.ndarray, so for the most part it can be used as an array, and arrays that fit the above description can be used in its place. Usually created by the factory functions: asBBox and fromPoints """ def __new__(subtype, data): """ Takes Data as an array. Data is any python sequence that can be turned into a 2x2 numpy array of floats:: [ [MinX, MinY ], [MaxX, MaxY ] ] You don't usually call this directly. BBox objects are created with the factory functions: asBBox and fromPoints """ arr = N.array(data, N.float) arr.shape = (2,2) if arr[0,0] > arr[1,0] or arr[0,1] > arr[1,1]: # note: zero sized BB OK. raise ValueError("BBox values not aligned: \n minimum values must be less that maximum values") return N.ndarray.__new__(subtype, shape=arr.shape, dtype=arr.dtype, buffer=arr) def Overlaps(self, BB): """ Overlap(BB): Tests if the given Bounding Box overlaps with this one. Returns True is the Bounding boxes overlap, False otherwise If they are just touching, returns True """ if N.isinf(self).all() or N.isinf(BB).all(): return True if ( (self[1,0] >= BB[0,0]) and (self[0,0] <= BB[1,0]) and (self[1,1] >= BB[0,1]) and (self[0,1] <= BB[1,1]) ): return True else: return False def Inside(self, BB): """ Inside(BB): Tests if the given Bounding Box is entirely inside this one. Returns True if it is entirely inside, or touching the border. Returns False otherwise """ if ( (BB[0,0] >= self[0,0]) and (BB[1,0] <= self[1,0]) and (BB[0,1] >= self[0,1]) and (BB[1,1] <= self[1,1]) ): return True else: return False def PointInside(self, Point): """ Inside(BB): Tests if the given Point is entirely inside this one. Returns True if it is entirely inside, or touching the border. Returns False otherwise Point is any length-2 sequence (tuple, list, array) or two numbers """ if Point[0] >= self[0,0] and \ Point[0] <= self[1,0] and \ Point[1] <= self[1,1] and \ Point[1] >= self[0,1]: return True else: return False def Merge(self, BB): """ Joins this bounding box with the one passed in, maybe making this one bigger """ if self.IsNull(): self[:] = BB elif N.isnan(BB).all(): ## BB may be a regular array, so I can't use IsNull pass else: if BB[0,0] < self[0,0]: self[0,0] = BB[0,0] if BB[0,1] < self[0,1]: self[0,1] = BB[0,1] if BB[1,0] > self[1,0]: self[1,0] = BB[1,0] if BB[1,1] > self[1,1]: self[1,1] = BB[1,1] return None def IsNull(self): return N.isnan(self).all() ## fixme: it would be nice to add setter, too. def _getLeft(self): return self[0,0] Left = property(_getLeft) def _getRight(self): return self[1,0] Right = property(_getRight) def _getBottom(self): return self[0,1] Bottom = property(_getBottom) def _getTop(self): return self[1,1] Top = property(_getTop) def _getWidth(self): return self[1,0] - self[0,0] Width = property(_getWidth) def _getHeight(self): return self[1,1] - self[0,1] Height = property(_getHeight) def _getCenter(self): return self.sum(0) / 2.0 Center = property(_getCenter) ### This could be used for a make BB from a bunch of BBs #~ def _getboundingbox(bboxarray): # lrk: added this #~ # returns the bounding box of a bunch of bounding boxes #~ upperleft = N.minimum.reduce(bboxarray[:,0]) #~ lowerright = N.maximum.reduce(bboxarray[:,1]) #~ return N.array((upperleft, lowerright), N.float) #~ _getboundingbox = staticmethod(_getboundingbox) ## Save the ndarray __eq__ for internal use. Array__eq__ = N.ndarray.__eq__ def __eq__(self, BB): """ __eq__(BB) The equality operator A == B if and only if all the entries are the same """ if self.IsNull() and N.isnan(BB).all(): ## BB may be a regular array, so I can't use IsNull return True else: return self.Array__eq__(BB).all() def asBBox(data): """ returns a BBox object. If object is a BBox, it is returned unaltered If object is a numpy array, a BBox object is returned that shares a view of the data with that array. The numpy array should be of the correct format: a 2x2 numpy array of floats:: [ [MinX, MinY ], [MaxX, MaxY ] ] """ if isinstance(data, BBox): return data arr = N.asarray(data, N.float) return N.ndarray.__new__(BBox, shape=arr.shape, dtype=arr.dtype, buffer=arr) def fromPoints(Points): """ fromPoints (Points). reruns the bounding box of the set of points in Points. Points can be any python object that can be turned into a numpy NX2 array of Floats. If a single point is passed in, a zero-size Bounding Box is returned. """ Points = N.asarray(Points, N.float).reshape(-1,2) arr = N.vstack( (Points.min(0), Points.max(0)) ) return N.ndarray.__new__(BBox, shape=arr.shape, dtype=arr.dtype, buffer=arr) def fromBBArray(BBarray): """ Builds a BBox object from an array of Bounding Boxes. The resulting Bounding Box encompases all the included BBs. The BBarray is in the shape: (Nx2x2) where BBarray[n] is a 2x2 array that represents a BBox """ #upperleft = N.minimum.reduce(BBarray[:,0]) #lowerright = N.maximum.reduce(BBarray[:,1]) # BBarray = N.asarray(BBarray, N.float).reshape(-1,2) # arr = N.vstack( (BBarray.min(0), BBarray.max(0)) ) BBarray = N.asarray(BBarray, N.float).reshape(-1,2,2) arr = N.vstack( (BBarray[:,0,:].min(0), BBarray[:,1,:].max(0)) ) return asBBox(arr) #return asBBox( (upperleft, lowerright) ) * 2 def NullBBox(): """ Returns a BBox object with all NaN entries. This represents a Null BB box; BB merged with it will return BB. Nothing is inside it. """ arr = N.array(((N.nan, N.nan),(N.nan, N.nan)), N.float) return N.ndarray.__new__(BBox, shape=arr.shape, dtype=arr.dtype, buffer=arr) def InfBBox(): """ Returns a BBox object with all -inf and inf entries """ arr = N.array(((-N.inf, -N.inf),(N.inf, N.inf)), N.float) return N.ndarray.__new__(BBox, shape=arr.shape, dtype=arr.dtype, buffer=arr) class RectBBox(BBox): """ subclass of a BBox that can be used for a rotated Rectangle contributed by MArco Oster (marco.oster@bioquant.uni-heidelberg.de) """ def __new__(self, data, edges=None): return BBox.__new__(self, data) def __init__(self, data, edges=None): """ assume edgepoints are ordered such you can walk along all edges with left rotation sense This may be: left-top left-bottom right-bottom right-top or any rotation. """ BBox.BBox(data) self.edges = np.asarray(edges) def ac_leftOf_ab(self, a, b, c): ab = np.array(b) - np.array(a) ac = np.array(c) - np.array(a) return (ac[0]*ab[1] - ac[1]*ab[0]) <= 0 def PointInside(self, point): for edge in xrange(4): if self.ac_leftOf_ab(self.edges[edge], self.edges[(edge+1)%4], point): continue else: return False return True