Adapting quantization offset in multiple description coding for error resilient video transmission

An adaptive approach to zooming-based control for uncertain systems with input quantization

“…x is bounded in view of Assumption 3. There will be a time instant, call it t 0 > 0, at which the following relation is true: (27) and as a consequence of (21), (25), e(t 0 ) ∈ I 1 (µ(t 0 )) ∩ B 1 (µ(t 0 )). Because e(t 0 ) ∈ B 1 (µ(t 0 )) and B 2 (µ) ⊂ B 1 (µ), we haveV ≤ 0 from (22).…”

“…In fact, dynamic offsets are often used in video encoding: in many H.264-based compression protocols the encoder is regulated on a frame level (offset) to obtain the number of bits that is very close to the allocated one [25]. Several mechanisms have been defined to tune the offset of the quantizer based on the previous frames [26], [27], so as to increase the resolution around similar frames close in time; similarly, our proposed quantizer increases the resolution around a dynamically changing offset in order to achieve asymptotic tracking.…”

A Mixed Layer Multiple Description Video Coding Scheme

“…In the side decoder, due to the spatial-temporal correlation within each frame (spatial) and between neighboring frames (temporal), the lost data can be estimated from the received data using error concealment techniques. In frequency domain MDC, the process of MDC generation is carried out in the DCT domain, such as DCT coefficient partitioning [13], [14], multiple description scalar quantizer (MDSQ) [15], [16], [18], quantization based MDC [17], [29], [41], and multiple description transform coding (MDTC) [19]- [23]. In [13], the coefficients larger than a certain threshold are replicated in all descriptions and the smaller ones are divided between descriptions.…”

“…In MDSQ, each source is assigned two indices, one for each description, such that with a given rate and side distortion, the maximum central quality is achieved. The methods which work with quantization decision levels can be placed in the category of MDSQ [18]. In quantization based MDC, the redundant data is the coarse quantized (lowerrate) representation of the primary data.…”

“…In the following parts, the optimal estimator is obtained for both CaseI and CaseII and then the performance of MLMDC with respect to the aforementioned distortions is studied. Ifx b is the quantized form of x b its probability distribution function can be found as follows: The distribution of x e in CaseI is obtained directly by the distribution of x b divided by the probability of the correspondinĝ x b as follows: (19) where P bN and P b0 were defined in (18). The distribution of x e is determined by the sign ofx b ifx b = 0, and the sign of x b ifx b = 0, as expressed in (19).…”