Some lossy compression algorithms

This repo contains two compression algorithms in python3:

• Color coding using huffman codes
• A custom jpeg implementation

Color coding

To perform color coding compression use color_code.py

usage: color_code.py [-h] [--verbose] {encode,decode} path_in path_out

positional arguments:
{encode,decode}  encode or decode an image
path_in          path to input
path_out         path for saving output

optional arguments:
-h, --help       show this help message and exit
--verbose        print encoding statistics

JPEG coding

To perform jpeg compression use jpeg_code.py

usage: jpeg_code.py [-h] [--verbose] {encode,decode} ... path_in path_out

positional arguments:
{encode,decode}  encode or decode an image
path_in          path to input
path_out         path for saving output

optional arguments:
-h, --help       show this help message and exit
--verbose        print encoding statistics

The JPEG encoding also allows for some optional parameters that control the level of compression. Higher compression results in a smaller file size but may lead to noticable artifacts.

usage: jpeg_code.py encode [-h] [--quant QUANT] [--sub_rows SUB_ROWS] [--sub_columns SUB_COLUMNS]

optional arguments:
-h, --help            show this help message and exit
--quant QUANT         level of quantization applied to the image. Lower values indicate less quantization.
                        Use integers ranging from -60 to 30 for good results
--sub_rows SUB_ROWS   chrominance subsampling in the direction of the rows
--sub_columns SUB_COLUMNS
                        chrominance subsampling in the direction of the columns

quant is the most important one. The default value is -45. You might want to try out a few quantization levels to see how much quantizaton affects your image.

Other interesting stuff

Besides the compression algorithms, there are also a couple of interesting things connected to the blog post. For example:

• Testing compression performance with picsum images
• Decomposing an image into its cosine transform components
• Projecting an image into a random orthogonal basis (and making a nice animation with it).