In order to improve quality of reconstructed frame in distributed video coding, a Deep Learning for Side Information(DL-SI) method was proposed. According to the fact that non-linear of pixel movement in a video sequence, the deep learning network was achieved and trained online. Therefore, the four key frames as the input block to the trained to predict the side information. Experiment results reveal that compared with the typical method IST-TDWZ, the peak signal noise ratio is improved by 2.4dB-3.7dB. This indicated that the quality of reconstructed video is significantly improved with deep learning.
展开▼