Algorithm 1 Face detection procedures |
Input: Original image
Output: The image with only the face
-
1:
Set the environment variables to match the features of the face.
-
2:
Adjust the image to fit the requirements.
-
3:
Pass the processed image into ResNet-101 and obtain the corresponding feature map.
-
4:
FPN corrects the size of the RoIs in the feature map.
-
5:
RPN classifies these RoIs and filters out the background, and BB regression corrects the BB of the RoI.
-
6:
Use RoI alignment to split the remaining RoIs into facial RoIs and non-facial RoIs.
-
7:
Use BB Regression to fix the BB of RoI again and generate the mask with FCN after classification.
-
8:
Keep the facial part after the non-facial part becomes black.
-
9:
End.
|