International standards such as ITU-T H.263 (H.261) and ISO MPEG use a hybr
id coding scheme based on the discrete cosine transform (DCT). Although the
DCT is an established technique for encoding interframe differences, it is
not an optimal transform for low-correlation interframe differences, since
its energy compaction performance largely depends on the correlation of th
e data. In this paper an effective technique for encoding interframe differ
ence is proposed and verified on an MPEG2 software encoder. The technique d
ecomposes the interframe difference into two sources in order to exploit th
eir improved statistical properties. The two-source interframe difference i
s encoded with a combined transform, which consists of the DCT and the Hada
mard transform (HT). Two-source decomposition increases the correlation of
the decomposed prediction error, and the DCT-HT scheme is shown to efficien
tly encode the decomposed interframe difference with increased performance.
Analytic and experimental comparison shows that the proposed DCT-HT scheme
gives significant objective and subjective gains over the conventional DCT
-based video encoding method. (C) 2001 Society of Photo-Optical Instrumenta
tion Engineers.