Optimization of multiple region quantizer for Laplacian source

Symbol coding of Laplacian distributed prediction residuals

“…When σ = τ = 1, the residual ε σ /τ =ε is integer-valued and the mapping (18) transforms into the Rice mapping (8). In the other extreme case of σ /τ → 0, the prediction x is not rounded at all.…”

“…Decoding: Given m, the decoder can compute M(ε σ /τ ) = jm + k. In Rice-Golomb coding, by checking whether M Rice (ε) is even or odd the decoder can decide which of the constituent functions in (8) was used by the encoder. Unlike the Rice mapping (8), the value of the both constituent functions in the residual mapping (18) can be even or odd.…”

“…Instead of using a single SQ, usage of two quantizers was suggested in [7]. This idea of using two quantizers was then extended in [8], where a multiple region quantizer for Laplacian sources has been analyzed. These schemes essentially focused on quantization of the Laplacian sources to lower the bit-rate at the expense of introducing some distortion.…”

“…Compression schemes for Laplacian sources, in the context of scalar quantization (SQ), have been proposed in [6][7][8]. A closed form equation to determine the optimal granular region for Laplacian sources was derived in [6].…”

Two forward adaptive dual-mode companding scalar quantizers for Gaussian source

“…Unlike [2] and [3], where the quantizer models are composed of SQs with an equal number of quantization levels, the models proposed in [4,5] are composed of SQs with an unequal number of quantization levels. To provide the highest possible quality of the quantized signal, given fixed bit rate, the support region thresholds of SQs observed in [2][3][4][5] have been optimized.…”