Compressive sampling for digital video signal compression involving dynamic sparsity

In this work, we consider a compressive sampling problem implemented in digital video compression. The compressive sensing method has a tremendous growth in recent researches. Inevitably, video compression field should explore alternatives in this novel approach to overcome some signal reconstruction drawbacks. We present the video signal representation and measurement based on dynamic sparsity of the source. Video sequences having high spatial and temporal redundancy are shown to take the most advantage out of compressive sampling technique. Firstly, we model the signal in a sparse domain using classic transforms such as DCT or wavelet. Then, in projection stage, scene dynamic is measured by the number of non significant coefficients. We use a quite small number C as threshold of these coefficients. In one group of picture (GOP), each frame is tested whether classified as sparse frame, hence undergoing compressive sampling; otherwise it would be processed by conventional sampling. The additional consideration of dynamic sparsity is expected to improve reconstruction quality, especially for video sequence with low redundancy property and low measurement rate. Furthermore, we compare the CVS results with those of MPEG-4 in terms of PSNR and compression ratio.