Image Processing Reference
In-Depth Information
the quality on the more salient pixels. This bit rate saving is obtained by decreasing
the quality of the less salient regions (the background) that are subject to lower
perceptual attention. However, this quality decrease remains limited and the PSNR
values measured on the entire images are very close with and without VAM. For the
“Musicians” sequence, the VAM approach achieves better performances even by
considering all the pixels for PSNR measurements. In this case, the VAM approach
improves the overall behavior of the encoder and provides more efficient informa-
tion to the rate-distortion control algorithm than the classical approach.
6.3.3.2 TIS Approach
With the temporal image signature (TIS) approach described in Sect.
6.2.3.1
, the
salient regions are sparse and the approach tends to highlight high frequency
regions like object borders. This behavior is particularly visible on the “Actors”
sequence (see Fig.
6.15
, bottom-left). However, for a block based video encoding
like H.264/AVC the quantization step cannot vary too much from a pixel block to
another due to the macroblock based (16
16 pixels) encoding and in order to limit
the blocking artifacts. Thus, to obtain the Qp offsets per macroblock, we apply
several post-processings:
• Median filtering (blurring) of the saliency map,
• Downscaling of the dense saliency map (each dimension divided by 16),
• Nonuniform quantization which aims at merging the more salient regions,
• Negative transformation to obtain Qp offsets.
Fig. 6.15 Saliency to quantization offset mapping with the TIS approach (
Top
: “Musicians”
sequence,
Bottom
: “Actors” sequence)
