Fast residue coding for lossless textual image compression

Lossless textual image compression based on pattern matching classically includes a "residue" coding step that refines an initially lossy reconstructed image to its lossless original form. This step is typically accomplished by arithmetically coding the predicted value for each lossless image pixel, based on the values of previously reconstructed nearby pixels in both the lossless image and its precursor lossy image. Our contribution describes background typical prediction (TPR-B), a fast method for residue coding based on "typical prediction" which permits the skipping of pixels to be arithmetically encoded; and non-symbol typical prediction (TPR-NS), an improved compression method for residue coding also based on "typical prediction". Experimental results are reported based on the residue coding method proposed in Howard's (see Proc. of '96 Data Compression Conf., Snowbird, Utah, p.210-19, 1996) SPM algorithm and the lossy images it generates when applied to eight CCITT bi-level test images. These results demonstrate that after lossy image coding, 88% of the lossless image pixels in the test set can be predicted using TPR-B and need not be residue coded at all. In terms of saved SPM arithmetic coding operations while residue coding, TPR-B achieves an average coding speed increase of 8 times. Using TPR-NS together with TPR-B increases the SPM residue coding compression ratios by an average of 11%.