Sparse flow adversarial model for robust image compression

摘要

Existing learned-based image compression methods have shown impressive performance. However, they rely mostly on the consistent distribution between training and test images, which reduces the robustness of the training model. In this paper, we propose a novel compression method called sparse flow adversarial model (SFAM). SFAM employs a deep generative framework to learn a reversible and stable mapping between image distributions, thus it can work in varied scenes for robust compression. The mapping explores the sparsity of the image by combining linear and nonlinear transformations, rather than extracting the features of a particular dataset as is the case with other learning-based methods. Moreover, a sparse adversarial map is introduced into SFAM, to constrain the SFAM to generate sparser features for efficient compression. Extensive experiments are performed on different datasets, in which the effectiveness and robustness of the proposed method are verified. Meanwhile, SFAM is trained only once and it can work well on three different datasets, which also prove the robustness of the proposed SFAM.