Higher precision range estimation for context‐based adaptive binary arithmetic coding

Double bit range estimation with eight estimators for CABAC in VVC

“…In the HEVC standard, CABAC is implemented as a state machine: corresponding to Figure 1 ≈ 0.949 (see our earlier work [6] for more details). It can be translated into a decreasing rate of updating of the symbolic probability values, which is a consequence of changing the value of the encoding bins.…”

“…Also, for estimators with slower adaptation rates, larger blocks are usually required. The encoded block in VTM maximum size is 2 6 , so in the range of x ≤ 6, we tend to restrict the slowest adaptation rate to r 7 less than the number of bits allocated to the fastest estimator P 0 (i.e. r 7 ≤ 20).…”

“…It follows that the size of the proposed multiplier is thus limited to 11 by 6 bits, where 11 is the number of bits representing the probability estimate and 8 is the number of bits representing the current interval range. Furthermore, it is proposed to use a positive integer 17 to subsample the probability estimate P into 2 6 states (e.g., similar to the 64 probability states of Figure 1), which can be further different for each context. By the same token, R LPS is also subsampled to 6 bits for the purpose of considering the preferred hardware implementation of a lesser number of bits multiplier.…”

“…These suggestions include the use of different distortion measures [22] and the use of different values of for 𝜆 [23][24][25]. More precisely, [6] proposed a more precise subdivision of the probability range, which is implemented as a binary fraction, with the multiplication of the descent rate that can be treated as a bitwise operation, limiting the possible values of 𝛼 to powers of two.…”

“…The coded video is segmented into rectangular blocks, and each block is predicted by internal or mutual prediction patterns. The difference/error of the resulting prediction is encoded using transform coding [5], which consists of an orthogonal transform, quantization of the transform coefficients, and entropy coding of the resulting quantization indices [6]. In addition, there are up to 35 intra‐prediction modes in HEVC and 68 intra‐prediction modes in VVC [7].…”