Low complexity distributed video coding using compressed sensing

Abstract

Compressive sensing (CS) is an efficient method to reconstruct sparse images with under-sampled data. In this method sensing and coding steps integrated to a one-step, low-complexity measurement acquisition system. In this paper, we use a Non-linear Conjugate Gradient (NLCG) algorithm to significantly increase the quality of reconstructed frames of video sequences. Our proposed framework divides sequence of a video to several groups of pictures (GOPs), where each GOP consisting of one key frame followed by two non-key frames. CS is then applied on each key and non-key frame with different sampling rates. For reconstruction final frames, NLCG algorithm was performed on each key frame with acceptable fidelity. To achieve desired quality on low-rate sampled non-key frames, NLCG modified using side information (SI) obtained from last two successive reconstructed key frames. Based on some performance measures such as SNR, PSNR, SSIM and RSE, our implementation results indicate that employing NLCG with Gaussian sampling matrix outperforms other methods in quality measures.