Summary: | 博士 === 國立臺灣大學 === 電機工程學研究所 === 84 === This thesis proposes two variations of a block truncation coding
(BTC) based system for data compression. The first is the visual
block pattern truncation coding (VBPTC) for still image
compression. The second is the minimized mean square error BTC
(MMBTC) for video data compression. VBPTC can reduce errors to
the least degree while catering to human visual characteristics.
Experimental results show that the algorithm increases the
compression ratio and removes the impulse noise. The MMBTC
algorithm minimizes mean square error (MSE). The video coding
system which combines MMBTC and conventional motion compensation
(MC) algorithm can eliminate both temporal and spatial
redundancy. BTC is a highly practical algorithm since its
computation is simple and it preserves block edges well. Unlike
the CCITT H.261 video coding system which relies on discrete
cosine transform (DCT), this newly introduced technique uses
BTC to preserve the block-oriented properties of predicted error
from motion compensation system. The four standard benchmark
images verify that the MMBTC algorithm is by far the best among
the various alternatives. The two standard benchmark sequences
also show that the MMBTC based system is much better than
conventional systems in terms of its implementation, complexity,
and accuracy.
|