Selective-PMO

H. Kojima et al., “Improved Probability Modeling for Lossless Image Coding Using Example Search and Adaptive Prediction”, IWAIT2022.

This implementation only supports 8-bpp grayscale images, raw PGM format.

Coding rates (bits per pixel)

Selective-PMO achieves the lowest coding rate (bpp) for most test images compared to state-of-the-art methods.

Image	Selective-PMO	PMO-GMM	MRP	Vanilc WLS D	TMW	Original
Camera	3.804	3.833	3.949	3.995	4.098	8
Couple	3.269	3.281	3.388	3.459	3.446	8
Noisesquare	5.274	5.296	5.270	5.159	5.542	8
Airplane	3.529	3.546	3.591	3.575	3.601	8
Baboon	5.611	5.698	5.663	5.678	5.738	8
Lena	4.215	4.237	4.280	4.246	4.300	8
Lennagrey	3.825	3.845	3.889	3.856	3.908	8
Peppers	4.161	4.176	4.199	4.187	4.251	8
Shapes	0.490	0.497	0.685	1.302	0.740	8
Balloon	2.573	2.584	2.579	2.626	2.649	8
Barb	3.708	3.733	3.815	3.815	4.084	8
Barb2	4.122	4.146	4.216	4.231	4.378	8
Goldhill	4.171	4.191	4.207	4.229	4.266	8
Average	3.750	3.774	3.826	3.874	3.923	8

PMO-GMM

Lossless Image Coding Exploiting Local and Non-local Information via Probability Model Optimization [paper]

MRP

Lossless coding using variable block-size adaptive prediction optimized for each image [paper] [source]

Vanilc WLS D

Probability Distribution Estimation for Autoregressive Pixel-Predictive Image Coding [paper] [source]

TMW

TMW - a New Method for Lossless Image Compression [paper] [source]

Original

All test images are from here.

Installation

1. Install the following requirements

   • CMake (tested with 3.17.0)
   • GCC (tested with 9.3.0) or Intel C++ Compiler (tested with 2021.2.0)

2. Generate a build file using CMake.

$ cd bin
$ cmake ..

3. Compile the project.

$ make

Usage

0. Print the usage of the options

$ ./pmo --help

1. Encode (.pgm to .pmo)

$ ./pmo -i /path/to/input/image.pgm -b /path/to/binary/file.pmo

2. Decode (.pmo to .pgm)

$ ./pmo -b /path/to/binary/file.pmo -o /path/to/output/image.pgm

3. Encode and Decode ( Check distortion-free )

$ ./pmo -i /path/to/input/image.pgm -b /path/to/binary/file.pmo -o /path/to/output/image.pgm

Authors

Hiroki KOJIMA (Maintainer), Diego FUJII