An image representation scheme by hybrid compressive sensing

Compressive sensing based image compression is a new paradigm shift in image representation and coding. The existing CS based image compression schemes are based on sparsity of the signals in transform domain. In this paper multiscale DWT is applied for sparse representation of the image. Instead of directly taking the measurements on sparse representation, statistical properties of low and high frequency subbands are studied. Based on the statistical properties of high and low band coefficients, a hybrid encoding scheme is proposed to encode low and high frequency subbands differently. At the decoder side, the combined bit stream is separated, and by fully exploiting the intra and inter scale correlation of multiscale DWT, different recovery algorithms are developed for low frequency and high frequency sub bands.