Image compression based on the transform domain downward conversion

In this paper, we focus on the design of a new block-based image compression by using our proposed transform domain downward conversion (TDDC). Applied directly on each 16×16 macro-block of pixels, this downward conversion is implemented through our proposed advanced padding technique such that a non-zero 8×8 coefficient block (thus down-sized) is generated only at the top-left corner in the transform domain, accompanied by zeros in other 75% positions. Consequently, a considerable bit-count saving can be achieved for the whole macro-block. In the meantime, 25% pixels reserved during the TDDC may be directly reconstructed from the down-sized coefficient block while the other 75% pixels that are not reserved during the TDDC will be reconstructed via the interpolation. Finally, this TDDC-based compression is used in conjunction with the JPEG baseline coding method (i.e., 1-out-of-2 selection) according to a rate-distortion optimization (RDO) based criterion. Experimental results show that our proposed compression scheme provides a significant quality gain as compared with the original JPEG baseline coding method and another super-resolution directed down-sampling (SRDDS) based compression scheme.