two problems.

[PapaMadeleine2022]

1. I do not understand why we need to erode the mask for dst_points here .

2.Without other op , simplely using seamlessClone op for blending two face image like cv2.seamlessClone(src_img, dst_img, src_mask, center, cv2.NORMAL_CLONE), I test 3 cases:

a) When I use all white mask for src_mask, there will be some white bright parts on the some edges of result face.
b) When I use the same-sized face white mask of src_img for src_mask, there will be some bigger white bright parts on the some edges of result face.
c) When I use the eroded face white mask of src_img for src_mask, there will be some black parts on the some edges of result face.

I can not find the causes and can not fix it. Can you give some advises?

Thanks.
Looking forward to your reply.

[wuhuikai]

1. Smaller face brings better visual quality.
2. Do you have any visual result?

[PapaMadeleine2022]

@wuhuikai Thanks for your reply.

Got 1. For q2:

This a simple demo code:

import cv2
import numpy as np

ori_img = cv2.imread("xxx/ori_img1.png")
img2 = cv2.imread("xxx/img2.png")

# Poisson Blending
h, w = img2.shape[:2]
center = (102 + int(w / 2), 120 + int(h / 2))
face_mask = 255 * np.ones(img2.shape, img2.dtype)  # Create an all white mask
result = cv2.seamlessClone(img2, ori_img, face_mask, center, cv2.NORMAL_CLONE)

After poisson Blending, I find that there is a little white part on the upper right corner of result face.

ori_img.png
im2.png
result.png

[wuhuikai]

That's the issue of Poisson Image Editing, please try another blending algos such as drag and drop.