A discrete cosine transform (DCT) level enhancement to Motion Picture
Experts Group (MPEG) video encoding is described that results in a more
concise bitstream than MPEG encoding without the enhancement. One degree
of freedom provided by the MPEG encoding specifications is whether a
frame- or field-based DCT operation will be used. In the field-based DCT
operations, luminance sub-blocks are built from even or odd rows of the
original image, which correspond to the top and bottom fields in
field-based video. This allows the encoder to take advantage of the
higher correlation between rows for the same field, especially in
field-based video with a high level of motion. In one embodiment, both
field- and frame-based DCT operations are performed and the results are
quantized. On a macroblock-by-macroblock basis, the option that results
in the fewest non-zero coefficients is selected and those coefficients
are used for run-time encoding.