如何在python中扭曲图像？

Question

How do i transform the image to make warp effect using pull warp like in the below image?

Left image is the input and right one is the output (warped version)

I read this article Image warping and found some hint. Now my question is that how do i calculate the offset ?

Answer 1

It looks like mission for scipy.ndimage.geometric_transform . Consider following example, using

1 2
3 4

as input 2D array

t = np.array([[1,2],[3,4]],dtype='uint8') #our input array
def func(x):
    return (x[1],x[0])
out = scipy.ndimage.geometric_transform(t,func)
print(out)

output:

[[1 3]
 [2 4]]

So what happend? For each cell in input, func with that cell position (tuple) as argument is calculated and its output used as new location for cell, remembering that indexes are starting at 0 and they are in y,x manner:

• 1 position is 0,0
• 2 position is 0,1
• 3 position is 1,0
• 4 position is 1,1

My example function simply switch y with x so

• 1 new position is 0,0
• 2 new position is 1,0
• 3 new position is 0,1
• 4 new position is 1,1

which indeed you could see in output.

Using cv2.imread function you can easily read image into numpy array, if you want solely transform grayscale image thats make things simpler as you would deal with 2D array rather than 3D array. Remember to use cv2.imread('someimage.bmp',0) (with 0 as second argument) when loading grayscale image.

Main challenge is creating the function: assuming grayscale its input is 2-tuple - position of "donor" pixel and output is 2-tuple - position of "recipient", note that this function do not have to cover whole output area - in case of output pixel missing "donor" it is interpolated.