Perceptually Optimized MPEG Compression of Synthetic Video Sequences
Enrico Masala, Davide Quaglia
Download full paper
The performance of the first and the second strategy
are tested with four synthetic video sequences created and animated by
3D Studio Max v.5.
An automatic object classifier was developed; it consider an object as
important if its distance from the point of view is
less than one fourth of the farthest object in the scene.
The importance of the objects is then used to decide the quantization
stepsize (QP) of the corresponding macroblocks.
For Strategy 1, we assigned a different QP value to important
(QPi) and less important (QPni)
macroblocks. For Strategy 2, macroblocks containing the edges of important
objects are quantized with a fixed QP value (QPb) which is
lower than QPi and QPni.
Frames are encoded by the TM5 MPEG 2 encoder which has been modified to
support the proposed techniques. The default GOP structure
consists of an I-frame followed by eleven P-frames.
Bitstreams encoded with the proposed techniques are
compared with model-unaware bitstreams encoded using
fixed values of the quantization stepsize. MPEG bitstreams and
sample frames are available below for perceptual evaluation.
Sequence Complex
Sequence Anibal
Sequence Dino
Sequence Dragon
