Adaptive lossless prediction based image compression

Predictive methods of image compression traditionally visit the pixels to be compressed in raster scan order, making a prediction for that pixel and storing the difference between the pixel and its prediction. We introduce a new predictive lossless compression method in which the order in which the pixels are visited is determined using a predictor based on previously known pixel values. This makes it possible to reconstruct the image without storing this path. In our tests on standard benchmark images; we show that our approach gives a significant improvement to row wise use of one or two dimensional predictors and gives results similar to or better than standard compression algorithms like median compression and JPG 2000.