What is this script?
Narrative Clip is a kind of life logging gadget. It takes a photo every 30 seconds (default setting). If we wear a Narrative Clip 10 hour a day, we get 1200 photos. This script make it possible to crop all photos square and combine them and make mp4 video, considering its rotation exif data and the necessity of being pixelated/blurred.
Requirement
- python3
- opencv
- ffmpeg
Download
Getting Started
brew install opencv3 --with-python3
brew install ffmpeg
set python script and face detection model file to below dir.
├── 2016
│ └── 05
│ └── 31
│ ├── 20160531_123827_000.jpg
│ ├── 20160531_123837_000.jpg
│ ├── :
│ ├── :
│ └── meta
│ ├── 20160531_123827_000.json
│ ├── 20160531_123827_000.snap
│ ├── 20160531_123837_000.json
│ ├── 20160531_123837_000.snap
│ ├── :
│ └── :
├── haarcascade_frontalface_default.xml <-- put file here
└── narrativeclip.py <-- put file here
execute the script by python narrativeclip.py {yyyy/MM/dd} {fps}
video.mp4 is output video file, and source images are components of its movie.
├── 2016
│ └── 05
│ └── 31
│ ├── 20160531_123827_000.jpg
│ ├── 20160531_123837_000.jpg
│ ├── :
│ ├── :
│ ├── meta
│ │ ├── 20160531_123827_000.json
│ │ ├── 20160531_123827_000.snap
│ │ ├── 20160531_123837_000.json
│ │ ├── 20160531_123837_000.snap
│ │ ├── :
│ │ └── :
│ ├── source00000.jpg <-- new file
│ ├── source00001.jpg <-- new file
│ ├── source00002.jpg <-- new file
│ ├── source00003.jpg <-- new file
│ └── video.mp4 <-- new file
├── haarcascade_frontalface_default.xml
└── narrativeclip.py
import glob
import PIL.Image as Image
import sys
import os
import re
import cv2
param = sys.argv
try:
pattern = r"\d{4}/\d{2}/\d{2}"
flag = re.match(pattern, param[1])
if len(param) < 3 or not flag or not int(param[2]):
raise Exception
except:
print("arg is {yyyy/MM/dd} {fps}")
sys.exit(1)
# rotate images by its exif, and crop them to square
files = glob.glob('{0}/[0-9]*.jpg'.format(param[1]))
convert_image = {
1: lambda img: img,
2: lambda img: img.transpose(Image.FLIP_LEFT_RIGHT),
3: lambda img: img.transpose(Image.ROTATE_180),
4: lambda img: img.transpose(Image.FLIP_TOP_BOTTOM),
5: lambda img: img.transpose(Image.FLIP_LEFT_RIGHT).transpose(Image.ROTATE_90),
6: lambda img: img.transpose(Image.ROTATE_270),
7: lambda img: img.transpose(Image.FLIP_LEFT_RIGHT).transpose(Image.ROTATE_270),
8: lambda img: img.transpose(Image.ROTATE_90),
}
for i, file in enumerate(files):
try:
print(file)
img = Image.open(file)
exif = img._getexif()
if exif is not None:
orientation = exif.get(0x112, 1)
img = convert_image[orientation](img)
w = img.size[0]
h = img.size[1]
if w > h:
l = h
diff = (w - h) / 2
s = img.crop((diff, 0, w - diff, h))
else:
l = w
diff = (h - w) / 2
s = img.crop((0, diff, w, h - diff))
counter = 0
pix = s.resize((50, 50)).load()
for j in range(50):
for k in range(50):
if pix[j, k][0] < 40 and pix[j, k][1] < 40 and pix[j, k][2] < 40: counter += 1
# remove black images
if counter < 2000: s.save('{0}/source{1:05d}.jpg'.format(param[1], i), "JPEG")
except Exception as e:
print("can not open " + file)
print(str(type(e)))
# pixelate each images if it has face
face_cascade = cv2.CascadeClassifier('haarcascade_frontalface_default.xml')
files = glob.glob('{0}/source*.jpg'.format(param[1]))
for file in files:
print(file + " mosaic")
img = cv2.imread(file)
result = cv2.imread(file)
gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
face = face_cascade.detectMultiScale(gray, scaleFactor=1.2, minNeighbors=1, minSize=(30, 30))
if 0 < len(face):
for (x, y, w, h) in face:
cut_img = img[y:y + h, x:x + w]
cut_face = cut_img.shape[:2][::-1]
# the size of being pixelated
cut_img = cv2.resize(cut_img, (int(cut_face[0] / 15), int(cut_face[0] / 15)))
cut_img = cv2.resize(cut_img, cut_face, interpolation=cv2.INTER_NEAREST)
result[y:y + h, x:x + w] = cut_img
cv2.imwrite(file, result)
cv2.destroyAllWindows()
os.system('ffmpeg -f image2 -r {1} -i {0}/source%05d.jpg -r {1} -an -vcodec libx264 -pix_fmt yuv420p {0}/video.mp4'.format(param[1],param[2]))