Said and Pearlman also have two lossy compression programs, Codetree and fastcode, using wavelet compression. codetree adds arithmetic entropy-coding, which improves the compression a little, but increases the time a lot. With these programs, you specify the desired number of bits per pixel, and whether the input data is smooth (used for optimizing the compression). However, since progcode also gives almost as good lossy compression, by decoding only part of the compressed file, we used that. Figure f:prograte shows the RMS (Root-Mean-Square) error of the compressed file for sizes ranging from 0.025 to 1.2 bpp, at which bit rate the reconstructed file is identical to the original.
Figure 3: Lossy Compression Adir512
We can now compare prograte to JPEG on the scaled data. For example, JPEG-75 has 0.45 bpp, and an RMS error of 6.2. However, prograte's RMS error at this reconstruction rate is 3.9. Codetree's RMS error at 0.45 bpp is 3.5. Alternatively, these programs can achieve an RMS error of 6.2 with a coding rate of only 0.25 bpp. Therefore they are almost twice as efficient as JPEG on the scaled file.