Saving image twice using PIL - python-2.7

Question

What happens if I save an image twice using PIL, with same image quality.
from PIL import Image
quality = 85
# Open original image and save
img = Image.open('image.jpg')
img.save('test1.jpg', format='JPEG', quality=quality)
# Open the saved image and save again with same quality
img = Image.open('test1.jpg')
img.save('test2.jpg', format='JPEG', quality=quality)
There is almost no difference in the image size or the image quality.
Can I assume that saving an image multiple times with same quality does not affect the actual image quality and that it is a safe to do so?
Also, if I save an image with 85% quality and then open and save with 95% quality, the image size becomes much larger. Does that mean PIL decompresses the image and compresses it again?

Answer 1

In most cases your test1.jpg and test2.jpg images will be slightly different. Meaning, a loss of information stored in test1.jpg will hapen after you open (decompress) and save it (compress again) with lossy JPEG compression.
In some cases however, opening and storing a JPEG image with same software will not introduce any changes.
Take a look at this example:
from PIL import Image
import os
import hashlib
def md5sum(fn):
hasher = hashlib.md5()
with open(fn, 'rb') as f:
hasher.update(f.read())
return hasher.hexdigest()
TMP_FILENAME = 'tmp.jpg'
orig = Image.open(INPUT_IMAGE_FILENAME)
orig.save(TMP_FILENAME) # first JPG compression, standard quality
d = set()
for i in range(10000):
# Compute file statistics
file_size = os.stat(TMP_FILENAME).st_size
md5 = md5sum(TMP_FILENAME)
print ('Step {}, file size = {}, md5sum = {}'.format(i, file_size, md5))
if md5 in d: break
d.add(md5)
# Decompress / compress
im = Image.open(TMP_FILENAME)
im.save(TMP_FILENAME, quality=95)
It will open and save a JPG file repeatedly until a cycle is found (meaning an opened image has exactly the same data as occurred before).
In my testing, it takes anywhere from 50 to 700 cycles to reach a steady state (when opening and saving image does not produce any loss). However, the final "steady" image is noticeably different from the original.
Image after first JPG compression:
Resulting "steady" image after 115 compress/decompress cycles:
Sample output:
Step 0, file size = 38103, md5sum = ea28705015fe6e12b927296c53b6d147
Step 1, file size = 71707, md5sum = f5366050780be7e9c52dd490e9e69316
...
Step 113, file size = 70050, md5sum = 966aabe454aa8ec4fd57875bab7733da
Step 114, file size = 70050, md5sum = 585ecdd66b138f76ffe58fe9db919ad7
Step 115, file size = 70050, md5sum = 585ecdd66b138f76ffe58fe9db919ad7
So even though I used a relatively high quality setting of 95, as you can see, multiple repeated compression/decompression made the image to lose its colors and sharpness. Even for quality setting of 100 the result will be very similar despite almost twice bigger file size.