Image Compression with Encoder-Decoder Matched Semantic Segmentation

Abstract

In recent years, layered image compression is demonstrated to be a promising\ndirection, which encodes a compact representation of the input image and apply\nan up-sampling network to reconstruct the image. To further improve the quality\nof the reconstructed image, some works transmit the semantic segment together\nwith the compressed image data. Consequently, the compression ratio is also\ndecreased because extra bits are required for transmitting the semantic\nsegment. To solve this problem, we propose a new layered image compression\nframework with encoder-decoder matched semantic segmentation (EDMS). And then,\nfollowed by the semantic segmentation, a special convolution neural network is\nused to enhance the inaccurate semantic segment. As a result, the accurate\nsemantic segment can be obtained in the decoder without requiring extra bits.\nThe experimental results show that the proposed EDMS framework can get up to\n35.31% BD-rate reduction over the HEVC-based (BPG) codec, 5% bitrate, and 24%\nencoding time saving compare to the state-of-the-art semantic-based image\ncodec.\n