VIDEO COMPRESSED SENSING AND RECONSTRUCTION METHOD AND APPARATUS BASED ON DEEP NEURAL NETWORK

摘要

Disclosed are a video compressed sensing and reconstruction method and apparatus based on a deep neural network. According to the method, a video signal is divided into a key frame and a non-key frame. The key frame is reconstructed by using an existing image reconstruction method. With regard to the non-key frame, the method innovatively proposes a special deep neural network to complete reconstruction. The neural network is composed of an adaptive sampling module, a multi-hypothesis prediction module and a residual error reconstruction module, and makes full use of the space-time correlation of the video signal to complete the sampling and reconstruction of the video signal, such that a low time complexity of an algorithm is guaranteed, while the reconstruction quality is improved. Therefore, the method of the present invention is suitable for a video sensing system having resources which are limited at a sampling end, and having high requirements regarding the reconstruction quality and real-time performance.