Encoding algorithm¶

JVol files are encoded using a simplified version of the JPEG algorithm, adapted for 3D images.

Color¶

For now, only single-channel images are supported. Therefore, no color space conversion or chroma subsampling is performed.

Block splitting¶

The array/volume of size $I \times J \times K$ is split into subvolumes/blocks of size $B \times B \times B$. If the array is not a multiple of $B$ along a dimension, it is padded using NumPy's reflect mode.

Here's an exaggerated illustration of reflection padding (in 2D, for visualization purposes):

Discrete cosine transform¶

TODO

Quantization¶

TODO

Entropy coding¶

TODO