The block is a matrix of 8x8 elements. They can be 8x8 adjacent luminance or chrominance samples, or the corresponding DCT coefficients.
At the block layer is performed the Discrete Cosine Transform.
Video decoding process at the block layer
Variable length decoding.
The bitstream codewords of the block are decoded to form a vector of quantised DCT coefficient.
Inverse scan.
The vector elements are put into a two-dimensional array, which is the block, following one of two possible patterns. The pattern is defined by the flag alternate_scan which is set at the picture layer. The scanning purpose is to optimize the entropy coding.
Inverse quantisation.
DCT coefficients are converted to their original range of values. See also dct quantisation.
Inverse DCT.
Eventually the IDCT is performed. Now the block elements represent either the image sample (Intra block) or the prediction-error (Non-Intra block).