H.264/AVC adopts macroblock-adaptive frame/field coding
mode (MBAFF) [1] to encode mixed regions in a picture.
Cause, tt is more efficient to compress each field separately the
region of motion existences in sequence. Whereas, the framQe
coding method is suitable for the non-moving regions in a
picture.
For producing thumbnail images in H.264/AVC only in
frame mode, a representative study was conducted by Kim et
al. [2]. Kim used a matrix operation for performing the intra
prediction in the transform domain.Thus, he adopted lookup
tables for compensating the errors, and directly extracted DC
values in the transform domain. However, the integer approximation
errors cannot be exactly compensated in transform
domain and so it is difficult for their method to control the
errors, and then induces mismatch errors between the encoder
and decoder [3]. As a result, it is urgent need to develop
a method for efficient generation of thumbnail images from
H.264/AVC compressed video adopting MBAFF mode.
In this work, we propose new algorithms to generate
thumbnail images from H.264/AVC intra-coded bit streams in
MBAFF mode. It not only radically eliminates the source of
the integer approximation errors, but also performs integer operations.
The developed method executes at least 2 times faster
than the full decoding method at frequently used bandwidths.