Do the dwt. You will get a vector of coefficients as output. Store only the first 42 coefficients for each block, discarding the rest. The result will be a file that is 42 / (8*8) = 62.625% as large as the original image, because the original image had one coefficient for every pixel, 64 coefficients for each 8*8 block.
When you want to decompress, load the coefficients back in, storing the 42 coefficients into each 8 * 8 block, and set the "missing" (64-42 = 22) coefficients to 0. Then do the inverse dwt.
