Full text (gzip-compressed postscript)
We present a method for progressive lossless compression of still grayscale images that combines the speed of our earlier FELICS method with the progressivity of our earlier MLP method. We use MLP's pyramid-based pixel sequence, and image and error modeling and coding based on that of FELICS. In addition, we introduce a new prefix code with some advantages over the previously used Golomb and Rice codes. Our new progressive method gives compression ratios and speeds similar to those of non-progressive FELICS and those of JPEG lossless mode, also a non-progressive method.
The image model in Progressive FELICS is based on a simple function of four nearby pixels. We select two of the four nearest known pixels, using the two with the middle (non-extreme) values. Then we code the pixel's intensity relative to the selected pixels, using single bits, adjusted binary codes, and simple prefix codes like Golomb codes, Rice codes, or the new family of prefix codes introduced here. We estimate the coding parameter adaptively for each context, the context being the absolute value of the difference of the predicting pixels; we adjust the adaptation statistics at the beginning of each level in the progressive pixel sequence.