Binary Probability Model for Learning Based Image Compression

Théo Ladune (1, 2) , Pierrick Philippe (2) , Wassim Hamidouche (1) , Lu Zhang (1) , Olivier Deforges (1)

1 IETR - Institut d'Électronique et des Technologies du numéRique
2 Orange Labs R&D [Rennes]

In this paper, we propose to enhance learned image compression systems with a richer probability model for the latent variables. Previous works model the latents with a Gaussian or a Laplace distribution. Inspired by binary arithmetic coding , we propose to signal the latents with three binary values and one integer, with different probability models. A relaxation method is designed to perform gradient-based training. The richer probability model results in a better entropy coding leading to lower rate. Experiments under the Challenge on Learned Image Compression (CLIC) test conditions demonstrate that this method achieves 18 % rate saving compared to Gaussian or Laplace models.

Keywords: Entropy Coding; Autoencoder; Convolutional Neural Network; Image Coding

Domains: Computer Science [cs]; Signal and Image Processing; Computer Science [cs]; Neural and Evolutionary Computing [cs.NE]; Computer Science [cs]; Artificial Intelligence [cs.AI]

Théo Ladune, Pierrick Philippe, Wassim Hamidouche, Lu Zhang, Olivier Deforges. Binary Probability Model for Learning Based Image Compression. ICASSP (International Conference on Acoustics, Speech, and Signal Processing) 2020, IEEE, May 2020, Barcelone, Spain. ⟨hal-02476067⟩