Converting to base64 changes numpy array

Question

I have a service which takes input a base64 image, converts it to numpy array, performs some operation on it and then converts it back to base64 image, which is sent as output. But the numpy array gets modified during the conversion to base64.

I have created an example to demonstrate the problem.

import base64
from io import BytesIO
import numpy as np
from PIL import Image
img = np.array([[[1, 2, 3], [1, 2, 3]], [[1, 2, 3], [1, 2, 3]]]
 ).astype(np.uint8)
print(img)
image_format = 'JPEG'
pil_img = Image.fromarray(img)
buff = BytesIO()
pil_img.save(buff, format=image_format)
encrypted_image = base64.b64encode(buff.getvalue()).decode('utf-8')
base64_decoded = base64.b64decode(encrypted_image)
img2 = Image.open(BytesIO(base64_decoded))
img2 = np.array(img2)
print(img2)

Output:

[[[1 2 3]
 [1 2 3]]
 [[1 2 3]
 [1 2 3]]]
[[[1 2 4]
 [1 2 4]]
 [[1 2 4]
 [1 2 4]]]

The last column of the img2 is different to that of img but I expected it to be the same.

Is the conversion to base64 wrong? If in what format should I take the input for my service?

• hobbs: JPEG is a lossy compression format.
• tdelaney: You get the same result if you remove the base64 stuff and just do buff.seek(0); img2 = Image.open(buff). Its not the encoding.
• aniketsharma00411: I checked after saving the array as PNG and the array is not getting modified now. So, it is the JPEG format that is causing the issue. Thanks @hobbs But is there any workaround to this or should I just stop using JPEG?
• Mark Setchell: JPEG was designed by the "Joint Photographic Experts Group" to do a good job of efficiently storing photographs. If you want accurately store scientific, technical, classification data you should use an appropriate format for that... PNG, TIFF, FITS, NetPBM or whatever depending on the scenario. So, it is fine to use JPEG, in the right circumstances.

Answer 1

Notice that you are not simply doing base64 encoding/decoding. Instead of x == base64.decode(base64.encode(x)) you are doing base64.decode(base64.encode(jpeg.deflate(jpeg.compress(x)))) == x

The part of base64 is bit accurate, but the part of JPEG compression will give only an approximation. In other words is not tru that jpeg.deflate(jpeg.compress(x)) always equals x.