This paper presents a new non-linear algorithm for image compression operating at different resolution levels and exploiting the binary versions of the intermediate images. The decimation factor that applies between the resolution levels along with the efficient coding of the binary signals allow considerable compression rates while maintaining the image quality. Lossless compression algorithms based on arithmetic coding are used to compress the binary files that are created during the image decomposition. The method was compared experimentally against the existing standard JPEG using the well known reference image “lenna” and it was shown that for similar compression rates it is blocking artifact free, offering at the same time higher peak SNR.