Abstract
In this paper, it is shown that there is useful method for improving PSNR of lossy Hadamard transform. This method cancelled mean component of the rounding errors which is generated at rounding process in transform, with offset added at rounding process in inverse transform.
As a result of improving PSNR of lossy Hadamard transform, partial of input data space is reversibly transformed. Input data space can be divided into two partial spaces. One is the space in which total sum of data is even. Another is the space in which total sum of data is odd. There are many vectors from which offset matrix can be generated such that one of tow partial spaces is reversibly transformed.
We propose new lossless 2D-Hadamard Transform which exchanges two offset matrix based on that the sum of data is odd or even. It is shown that entropy of the lossless transformed coefficients is scarcely affected by selection of offset matrix.
