In this paper, they propose a deep semantic segmentation-based layered image compression (DSSLIC) framework in which the semantic segmentation map of the input image is obtained and encoded as the base layer of the bit-stream. A compact representation of the input image is also generated and encoded as the first enhancement layer. The segmentation map and the compact version of the image are then employed to obtain a coarse reconstruction of the image. The residual between the input and the coarse reconstruction is additionally encoding as another enhancement layer.

Network Details

  • FLIF enc/dec: state-of-the-art lossless image codec.
  • BPG enc/dec: state-of-the-art lossy coding.
  • CompNet for generating a low-dimensional version \(c\) of the original image.
  • SegNet for generating the semantic segmentation map \(s\) of the input image \(x\).
  • FineNet is trained to learn the missing fine information of the up-sampled version of \(c\) concerning the input image.


Other application

Since semantic segmentation map includes in the bit-stream, the proposed scheme can facilitate many other tasks such as image search and object-based adaptive image compression.