Face detection using RetinaFace and image cropping to the desired format @300dpi.

In case the models are removed from the original repo, or there is a problem downloading the models, they are also available here: Google Drive

FIPU face detects possible errors in the image:

• picture taken to close to the camera (too_close)
• head tilted to the side (head_tilted)
• head turned to the side (head_turned)
• no faces in the image (no_face)
• multiple faces in the image (too_many_faces)
• image is too blurry (image_blurry)
• image background is not white (non_white_bg)
• eyes, nose or mouth not detected (no_landmarks)

All errors (and appropriate messages) are stored in ImageException.py

If none of the errors are detected on the image, the face is cropped and resize'd to the desired format.

Supported formats:

• x_220x300 (18.6x25.4)
• x_300x300 (25.4x25.4)
• 30x35_11Plus
• 30x35_11
• 35x45_11Plus
• 35x45_11

1. Uploading a file through a form (param name img)
2. Uploading a file in base64 encoding using form/json (param name img64)
3. Uploading a file encoded as bytes using form/json (param name img_bytes) (iso-8859-1)
4. Specifying output encoding using form/json (param name resp_enc)
5. Specifying image format(s) using form/json (param name img_sizes) (string, list, dictionary)

Request example:

{"img64": "/9/dsjdias...", # Original image in base64 format
 "img_sizes": {"img_big": "x_300x300", "img_small": "x_220x300"} # Or "x_220x300", or ["x_220x300", "x_300x300"]
}

Response returns two types of json objects:

• When error occurs

{"img": "/9/dsjdias....", # Image with errors displayed in it
 "errors": [{'error_code': 'err1', 'message': 'Detail message error 1'},
            {'error_code': 'err2', 'message': 'Detail message error 2'}]}

• When image processing completed successfully

If only one format was asked

{"img": "/9/dsjdias...."} # Processed image in base64/bytes(iso-8859-1) format 

If multiple formats were asked (list, or tuple)

{
 "img0": "/9/dsjdias....", # Processed image in base64/bytes(iso-8859-1) format 
 "img1": "/9/dsjdias...." # Processed image in base64/bytes(iso-8859-1) format
}

If multiple formats were asked (dictionary)

{
 "img_big": "/9/dsjdias....", # Processed image in base64/bytes(iso-8859-1) format using the key form the request 
 "img_small": "/9/dsjdias...."# Processed image in base64/bytes(iso-8859-1) format using the key form the request
}

x	30x35@300dpi	35x45@300dpi

pip install -r requirements.txt

Place your image in imgs folder and run test.py. New image would be placed under the same name in imgs/new folder. If the validation fails the image, with the errors on the image, will be placed in imgs/draw. The error codes of the exceptions will also bi printed in the console.

Create a .env file with following content (adjustment needed):

FLASK_ENV=development
BUGSNAG_API_KEY=

from fipu_face.fipu_face import *
import os


def do_detect(stream_path):
    print(stream_path)
    frame = cv2.imread('imgs/' + stream_path)
    try:
        frame = detect(frame)
        cv2.imwrite('imgs/new/' + stream_path.replace('.jpg', '.jpg'), frame, [cv2.IMWRITE_JPEG_QUALITY, JPEG_QUALITY])
    except ImageException as e:
        cv2.imwrite('imgs/draw/' + stream_path, e.image, [cv2.IMWRITE_JPEG_QUALITY, JPEG_QUALITY])
        print(e.get_error_codes())


if __name__ == '__main__':
    for i in sorted([f for f in os.listdir('imgs/') if f.split('.')[-1] in ['jpg', 'jpeg', 'png'] ]):
        do_detect(i)

You can also check how your face is cropped, and what errors are displayed, in real time by running tests/camera.py

• Image cropping for official documents (in Croatia)

• Emotion detection - government document should have neutral emotion

• Check for face obstacles (eg. glasses, mask, hand in front of face)

• More accurate background color detection

• Portrait matting - using PyMatting?

• Api unit testing

• Fipu face feature testing

  • Detection testing
  • Background testing
  • Cropping testing
  • Blur testing
  • Face alignment testing

• RetinaFace - Face detection
• Portrait Segmentation - Non white background detection
• FER - Facial expression recognition (in progress)
• Real-time Glasses Detection - Glasses detection (in progress)
• Flask - API