It is well known that there are a lot of noise data in many digital images. In generally speaking

the embedded noise data in an image are not only harmful to view the image

but also there is less correlation among noise data and original image data. So the noise data are hard to be compressed by general compression methods based on predictive or schotistic coding. Therefore a new method to achieve nearly lossless compression of an image is proposed in this paper. In the first step in the proposed method the noise data in an image are eliminated based on a noise model

and then the resulted image is as better as the original one for the usages. In the second step

the image is encoded by using a region adaptive subband compression algorithm without loss any data. Obviously

after decoding

the reconstruction image is a nearly lossless image without harmful to the usages. Coding time of the proposed algorithm is affordable thanks to fast convergence of the algorithm. Coding could always be performed in real time. The experimental result shows that the compression scheme provides impressive performance such as high signal to ratio and high compression ratio. According to the theory of the algorithm

the noise elimination algorithm based on noise model can also be extended to the other compression algorithms such as DCT

DPCM

JPEG

SPIHT and MPEG.