The recovery of a full resolution color image from a color filter array like the Bayer pattern is commonly regarded
as an interpolation problem for the missing color components. But it may equivalently be viewed as the problem
of channel separation from a frequency multiplex of the color components. By using linear band-pass filters in
a locally adaptive manner, this latter view has earlier been successfully approached, providing state-of-the-art
performance in demosaicking. In this paper, we address remaining shortcomings of this frequency domain method
and discuss a locally adaptive restoration filter. By implementing restoration as an extension of the bilateral
filter, reasonable complexity of the method can be sustained while being able to improve resulting image quality
by up to more than 1dB.
A locally adaptive up-sampling method that improves the efficiency of a spatially scalable representation of
images in a spatial pyramid is presented. While linear methods use a globally optimized up-sampling filter
design, the method presented locally switches between enhancement of significant structures and smoothing of
flat regions that are dominated by noise. It is based on a locally adaptive Wiener filter expression that can
be implemented by the bilateral filter. The performance of the method is assessed in a scenario resembling its
possible use in MPEG's and ITU-T's joint current activity on scalable video coding (SVC).
Compression of digitized video highly depends on, and varies with, the signal to be compressed. The relation of quantizer, distortion and rate needs to be modeled when control in a system involving video compression is asked for. An experimental analysis of the optimized video codec AVC of ITU-T and MPEG shows that its rate behavior can be modeled accurately enough to predict bit-rate on macroblock-level. Given information of allocated rates from a pre-encoding analysis step, the bit-rate profile at any different quantizer setting for the video can be predicted. Experiments, comparing predicted rate against actual encoding rate show good performance of the model.These reflect the performance of the model in rate-control schemes. A simple
example pre-analysis rate-control based on the model will determine beforehand a possible rate-profile that the actual encoding should be able to follow with small quality variations. For a signal of varying complexity, a varying number of bits will be used to obtain constant quality. Such variations are limited by peak bit-rates and buffer-sizes that are defined by hypothetical reference decoders in AVC.
State-of-the-art video coders feature a high level of adaptivity to meet the properties of their input signals. Video signal statistics are apt to sometimes discontinuously vary in all three dimensions:
spatially and temporally. Consequently for video compression, different coding tools are employed at different coordinates of the signal to adaptively and maximally reduce redundancy. Since the decision for a certain tool and for the apropriate parameters is highly signal-dependent, a video coder forms a non-linear system and optimization is not trivial to perform. Lagrange Rate/Distortion-Optimization (RDO) has become an important tool for video encoding. It has achieved high gains in coding efficiency when applied to
the independent encoding of macroblocks (MBs). For a chosen
quantizer, bit rate is allocated to each MB's coefficients and prediction parameters subject to minimization of a Lagrange cost function of rate and distortion. In effect, the cheaper option is chosen to either use a more precise prediction, i.e. motion vectors and block tiling, or to spend more bit rate on the coefficients.
Often, the video coding objective is that of constant quality. This is approximately achieved by a constant quantizer for all MBs and RDO within each MB. Due to the recursive structure of hybrid video coders though, MBs are dependent on each other, demanding for dependent RDO. This has early been formulated, but is hardly solvable due to the high dimensionality of this problem. In the following, possible simplified ways to take dependencies into account are explored. The variations in all three dimensions of the signal properties are differentiated and simplified models used for a dependent optimization.