Recently, Compressive Sensing (CS) has emerged as a more efficient sampling method for sparse signals. Comparing to
the traditional Nyquist-Shannon sampling theory, CS provides a great reduction of sampling rate, power consumption, and
computational complexity to acquire and represent sparse signals. In this paper, we propose a new block based image/video
compression scheme, which uses CS to improve coding efficiency. In the traditional lossy coding schemes, such as JPEG
and H.264, the dominant coding error comes from scalar quantization. The CS recovery procedure can help mitigating the
quantization error in the decoding process. We use rate distortion optimization (RDO) for mode selection (MS) between
the traditional inverse DCT transform and projection onto convex sets (POCS) algorithm. In our experiment, the new image
compression method is able to achieve up to 1 dB gain over standard JPEG.