Report on JP3D compression of solar data-cubes

The performance of the recent JPEG2000 Part 10 standard, known as JP3D, is evaluated for the lossless compression of several types of solar data-cubes.

We explore the differences in:

a) The compressibility of broad-band or narrow-band; I or V stokes profiles b) Compressing data in x,y,lambda packages at different times or data in x,y,t packages of different wavelength c) Compressing a single large data-cube or several smaller data-cubes d) Compressing data which is under-sampled or super-sampled wrt the diffraction cut-off.

The performance of the 3D compression varies with the data-type: G-band are the most compressible ( 3 bits per voxel), while Stokes I are the less compressible ( 7 bits per voxel). The gain against a 2D compression varies from -50% to -30% of the original data volume. The gain in ordering data in x,y,lambda,time or x,y,time, lambda is apparently negligible. The spatial correlation present in data which are super-sampled wrt the telescope cut-off frequency leads to a +33% in compression rate (but on a 4x data!). As expected, JP3D automatically compress the same data-cube to the same BPV, whether it is provided as 16bit or 32bit data.

Here are the data which have been used for this analysis