A python package for accessing the internal variables of JPEG file format such as DCT coefficients and quantization tables.

It is recommended to install by compiling yourself. The installation process includes compiling C/C++ source codes.

python setup.py install

You can also use the pre-compiled wheels (If some errors occur, try compiling the package using the above command).

On Linux:

pip install dist/jpegio-x.x.x-cp3x-cp3x-linux_<your architecture>.whl

On Windows 10 x64:

pip install dist/jpegio-x.x.x-cp3x-cp3x-win_amd64.whl

On macOS x64:

pip install dist/jpegio-x.x.x-cp3x-cp3x-macosx_<version>_x86_64.whl

You can use the following command for making a wheel for your own architecture such as x64 or ppc64le.

python setup.py bdist_wheel

The cooked wheel files are located at dist directory.

This package requires other packages.

• Cython
• NumPy

import jpegio as jio

jpeg = jio.read("image.jpg")
coef_array = jpeg.coef_arrays[0]  
quant_tbl = jpeg.quant_tables[0]

# Modifying jpeg.coef_arrays...
# Modifying jpeg.quant_tables...

jio.write(jpeg, "image_modified.jpg")

• coef_arrays is a list of numpy.ndarray objects that represent DCT coefficients of YCbCr channels in JPEG.
• quant_tables is a list of numpy.ndarray objects that represent the quantization tables in JPEG.

You can also utilize other variables (one of the simplest ways for finding them is to use dir(img)). The names of member variables have been determined following the convention of libjpeg.

• The core parts of this package, implemented in C/C++, are adopted from the souce codes of Jessica Fridrich's laboratory.
• The functionality of libjpeg is borrowed from IJG and libjpeg-turbo.

Apache License 2.0