给定一个向量场 (dx, dy)，将位置 (Row, Col) 处的矩阵值移动到新位置 (Row + dx, Column + dy)

Question

Given a matrix

[   a   b   -  ]
[   -   e   f  ]
[   g   h   -  ]

where, for the sake of demonstration, - denotes a zero entry.

We also work with a vector field

[   (0,1)   (0,1)   (0,0)  ]
[   (0,0)  (0,-1)  (0,-1)  ]
[   (0,1)   (0,1)   (0,0)  ]

where each tuple specifies how many (rows, columns) to move the corresponding element in the matrix. What is a Pythonic/efficient way to move each element by its corresponding vector to achieve:

[   -   a   b  ]
[   e   f   -  ]
[   -   g   h  ]

This was inspired by a coregistration problem, but I haven't found an elegant solution to this problem besides looping through element wise. I'm new to image processing, and also programming in Python - what is an efficient/accepted way to do this?

Answer 1

This can be done using np.add.at :

A = np.array([["a","b",""],["","c","d"],["e","f",""]])
l,n,r = [[0,-1],[0,0],[0,1]]
B = np.array([[r,r,n],[n,l,l],[r,r,n]])

out = np.zeros_like(A)
i,j = np.ogrid[:3,:3]
np.add.at(out.view('u4'),(i+B[...,0],j+B[...,1]),A.view('u4'))

out
# array([['', 'a', 'b'],
#        ['c', 'd', ''],
#        ['', 'e', 'f']], dtype='<U1')