ternaus / base64toimageconverters

Library for converting from RGB / GrayScale image to base64 and back.

Library for converting RGB / Grayscale numpy images from to base64 and back.

Installation

pip install -U image_to_base_64

Conversion

Takes numpy.array or PIL.Image as input

RGB to base 64

base64 = rgb2base64(rgb_image, image_format)

where image format is JPEG, PNG

Grayscale to base 64

base64 = grayscale2base64(grayscale_image)

Base64 to RGB image

output_type = "cv2" # or "PIL"
rgb_image = base64_to_rgb(base64, output_type)

Base64 to Grayscale image

output_type = "cv2" # or "PIL"
grayscale_image = base64_to_grayscale(base64, output_type)

Issues

For some reason I cannot convert RGB image to JPEG representation in base 64 and back without losses. => test only for PNG and not JPEG

base64toimageconverters's Issues

About the JPEG encoding, decoding

Referring to the issue in the README, saving the image as JPEG will inevitably apply some level of compression. So decoding the image to retrieve the exact pixel intensities of the original image may not be possible unless using lossless formats like PNG

However, I found a little parameter of PIL save subsampling while doing the JPEG compression which tries to maintain the pixel intensity, still incurring some loss but not as big as without this parameter.

Given codebase without subsampling

Original implementation: https://github.com/ternaus/base64ToImageConverters/blob/main/image2base64/converters.py
Line number: 12
format: 'JPEG', quality=100

im.save(buffered, format=image_format, quality=quality)

RGB Image:

Converted Image:

Mean difference of the assertion test: 125.74196689386562

With Subsampling parameter set to 0

im.save(buffered, format=image_format, quality=quality, subsampling=0)

RGB Image:

Converted Image:

Mean difference of the assertion test: 53.83414475819539

The visual as well as numerical difference in clear. JPEG will definitely lead to losses. But subsampling=0 can preserve more details.