Distributed compressed video sensing based on recursive least square dictionary learning

摘要

In this paper, we propose a method for distributed compressed video sensing (DCVS) based on dictionary learning. The proposed method divides the video sequences into group of pictures (GOP). Each GOP includes a key-frame following by a CS-frame. Compressed sensing (CS) is used to exploit spatial redundancy of frames. At the encoder side Key-frames are sampled using random projection methods. To acquire much sparser version of CS-frames, a basis extracted from CS-frame itself, using dictionary learning approach and used as a sparsifying basis. Sampling rate for key-frames and CS-frames are respectively adjusted to 0.5 and 0.25. At decoder side each frame reconstruction formulated as an l1-minimization problem. For each CS-frame, motion compensation interpolation method is applied on previous reconstructed key-frames to generate side information (SI). A dictionary is learned from SI and is used as a basis function in order to compensate low sample rate of CS-frames based of recursive least square dictionary learning algorithm (RLS-DLA). The results comparison with iterative least square dictionary learning algorithm (ILS-DLA) and K-SVD algorithm shows that the proposed method performs better than dictionaries learned by other methods.