In [1]:

    

import cv2
import numpy as np


    

In [2]:

    

src = cv2.imread('data/src/lena.jpg')
dst = cv2.imread('data/src/rocket.jpg')


    

In [3]:

    

src_pts = [[100, 80], [150, 200], [300, 20]]
dst_pts = [[280, 120], [320, 300], [400, 150]]


    

In [4]:

    

src_mark = src.copy()

for pt in src_pts:
    cv2.drawMarker(src_mark, tuple(pt), (0, 255, 0), thickness=4)

cv2.imwrite('data/dst/opencv_warp_src_mark.jpg', src_mark)


    

    
Out[4]:





True

In [5]:

    

dst_mark = dst.copy()

for pt in dst_pts:
    cv2.drawMarker(dst_mark, tuple(pt), (0, 255, 0), thickness=4)

cv2.imwrite('data/dst/opencv_warp_dst_mark.jpg', dst_mark)


    

    
Out[5]:





True

In [6]:

    

src_pts_arr = np.array(src_pts, dtype=np.float32)
dst_pts_arr = np.array(dst_pts, dtype=np.float32)


    

In [7]:

    

src_rect = cv2.boundingRect(src_pts_arr)
dst_rect = cv2.boundingRect(dst_pts_arr)


    

In [8]:

    

print(src_rect)


    

    




(100, 20, 201, 181)

In [9]:

    

print(dst_rect)


    

    




(280, 120, 121, 181)

In [10]:

    

src_crop = src[src_rect[1]:src_rect[1] + src_rect[3], src_rect[0]:src_rect[0] + src_rect[2]]
dst_crop = dst[dst_rect[1]:dst_rect[1] + dst_rect[3], dst_rect[0]:dst_rect[0] + dst_rect[2]]


    

In [11]:

    

src_pts_crop = src_pts_arr - src_rect[:2]
dst_pts_crop = dst_pts_arr - dst_rect[:2]


    

In [12]:

    

print(src_pts_crop)


    

    




[[  0.  60.]
 [ 50. 180.]
 [200.   0.]]

In [13]:

    

print(dst_pts_crop)


    

    




[[  0.   0.]
 [ 40. 180.]
 [120.  30.]]

In [14]:

    

src_crop_mark = src_crop.copy()

for pt in src_pts_crop.astype(np.int):
    cv2.drawMarker(src_crop_mark, tuple(pt), (0, 255, 0), thickness=4)

cv2.imwrite('data/dst/opencv_warp_src_crop_mark.jpg', src_crop_mark)


    

    
Out[14]:





True

In [15]:

    

dst_crop_mark = dst_crop.copy()

for pt in dst_pts_crop.astype(np.int):
    cv2.drawMarker(dst_crop_mark, tuple(pt), (0, 255, 0), thickness=4)

cv2.imwrite('data/dst/opencv_warp_dst_crop_mark.jpg', dst_crop_mark)


    

    
Out[15]:





True

In [16]:

    

mat = cv2.getAffineTransform(src_pts_crop.astype(np.float32), dst_pts_crop.astype(np.float32))
affine_img = cv2.warpAffine(src_crop, mat, tuple(dst_rect[2:]))

cv2.imwrite('data/dst/opencv_warp_affine_crop.jpg', affine_img)


    

    
Out[16]:





True

In [17]:

    

mask = np.zeros_like(dst_crop, dtype=np.float32)
cv2.fillConvexPoly(mask, dst_pts_crop.astype(np.int), (1.0, 1.0, 1.0), cv2.LINE_AA)

dst_crop_merge = affine_img * mask + dst_crop * (1 - mask)

cv2.imwrite('data/dst/opencv_warp_affine_crop_merge.jpg', dst_crop_merge)


    

    
Out[17]:





True

In [18]:

    

dst[dst_rect[1]:dst_rect[1] + dst_rect[3], dst_rect[0]:dst_rect[0] + dst_rect[2]] = dst_crop_merge

cv2.imwrite('data/dst/opencv_warp_dst_result.jpg', dst)


    

    
Out[18]:





True