Video Compressive Sensing with Side Information

We report a temporal compressive imaging system with side information. In addition to a coded measurement used to play the pivot role to reconstruct high speed video frames, we employ a second channel to capture the blurry image as side information to improve the reconstruction quality. These two channels are operating at the same low frame rate to capture the high speed scene. In the first channel, each high-speed video frame is modulated by a physical shifting to form a coded image and these coded images are superimposed to construct the measurement captured by the sensors. The second channel is directly capturing the summarization of these high-speed frames. Following this, a joint reconstruction algorithm is developed to recover the high-speed videos from the two measurements. We demonstrate a 4x temporal compression imaging system by dividing the detector into two parallel views and thus using a single detector to capture the two channels simultaneously. Simulation and experimental results demonstrate the improved performance of the reconstructed videos aided by the side information