DCT coefficient decoder

Python wrapper for decoding DCT coefficients from JPEG-compressed images.

Note: This is research code and has not been designed for use in production. Tested with Python 3.5 and libjpeg 9a.

Prerequisites

• Header files for building Python extensions

apt-get install python3-dev

• You need to have libjpeg installed. If you don't have a global installation, you can download libjpeg from here and install it by following the build instructions. Please update the include and lib paths in setup.py such that they point to your installation.

libjpeg_include_dir = "path/to/libjpeg/build/include"
libjpeg_lib_dir = "path/to/libjpeg/build/lib"

• Python dependencies

pip install Cython numpy

Compile

Global installation

pip install .

This will automatically install the python dependencies.

Local installation

To compile the extension for use in the project's root directory, run the following:

python setup.py build_ext --inplace

This will place a shared object in the current working directory. For use from another directory, make sure to include this module in your PYTHONPATH.

Usage

from decoder import PyCoefficientDecoder 

filename = "image.jpg"

# Load and decode image
d = PyCoefficientDecoder(filename) 

# Access image width
d.image_width

# Read (quantized) DCT coefficients of luminance channel
d.get_dct_coefficients(0)

# Read quantization table for luminance channel
d.get_quantization_table(0)

Example how to vary the DCT implementation. Supported options are JDCT_ISLOW (default), JDCT_IFAST, JDCT_FLOAT.

from decoder import PyCoefficientDecoder, JDctMethod
filename = "image.jpg"
d = PyCoefficientDecoder(filename, dct_method=JDctMethod.JDCT_FLOAT)
img = d.get_decompressed_image()

Known issues

• Attempting to read uncompressed image after reading DCT coefficients results in Improper call to JPEG library in state 210. Workaround: Initialize another instance of PyCoefficientDecoder.

Misc

1. Another way to quickly read an image's quantization tables:

djpeg -verbose -verbose image.jpg > /dev/null

2. A visual explanation of the Discrete Cosine Transform (DCT) is given in this notebook.