Lossless encoding and decoding of digital data
View Patent ↗The disclosure relates to encoding and decoding of digital data, and in particular to lossless arithmetic encoding and decoding of digital data representing audio, image or video data. A probability density function used for lossless arithmetic encoding of digital data is controlled by employing one or more parameters that changes over the set of data to be encoded. A parametric model in the form of an envelope function describes the spread of quantization indices derived from the data in a transform domain. By transmitting the one or more parameters together with the arithmetically encoded data, a receiving decoder may decode the data by exploiting the same parametric model as used by the encoder.
1. A method of lossless encoding of a block of quantization indices in a transform domain for transmission as a bitstream, the method including:
modeling the spread of the quantization indices with an envelope function which is described by at least one envelope parameter;
for each quantization index:
recreating an upper boundary and a lower boundary of a quantization region corresponding to the index;
determining a respective value of the at least one envelope parameter value at a position of the envelope function that corresponds to the index; and
determining upper and lower probability values for the index by multiplying said upper and lower boundaries with the inverse of the determined at least one parameter value; and
arithmetically encoding the upper and lower probability values for the index,
encoding the at least one envelope parameter describing the envelope function; and
providing a bitstream for transmission to a decoder, which bitstream includes the arithmetically encoded quantization indices and the encoded at least one envelope parameter.
2. The method of claim 1 , wherein the quantization indices represent an audio signal, the method including performing a time-to-frequency transformation of digitized audio samples to obtain frequency coefficients and then quantizing the frequency coefficients, wherein said step of modeling the spread of the quantization indices comprises modeling the spread of the quantized frequency coefficients.
3. The method of claim 1 , wherein the quantization indices represent a pixel based image, the method including performing a space-to-frequency transformation of digitized pixel data to obtain spatial frequency coefficients and then quantizing the spatial frequency coefficients, wherein said step of modeling the spread of the quantization indices comprises modeling the spread of quantized spatial frequency coefficients.
4. The method of claim 1 , wherein said upper and lower probability values are values of a cumulative probability density function.
5. The method of claim 4 , wherein the probability model of the cumulative probability density function is based on a distribution included in the set of: Gaussian probability distribution, a logistic distribution and a Laplace distribution.
6. The method of claim 1 , wherein said at least one envelope parameter include an envelope scale parameter.
7. The method of claim 6 , wherein the envelope model is an Auto-Regressive, AR, model and the envelope parameters include AR coefficients.
8. A method of lossless decoding of a block of quantization indices in a transform domain received as a bitstream, the method including:
separating a received bitstream into an arithmetic bitstream and a description of at least one encoded envelope parameter;
decoding the at least one envelope parameter;
determining an envelope function based on the at least one envelope parameter, the envelope function modeling the spread of coded quantization indices;
for each quantization index to be decoded:
arithmetically decoding a pair of upper and lower probability values from the arithmetic bitsream;
tentatively coding a quantization index by recreating an upper boundary and a lower boundary of a quantization region corresponding to the index, determining a respective value of the at least one envelope parameter value at a position of the envelope function that corresponds to the index, and determining upper and lower probability values for the index by multiplying said upper and lower boundaries with the inverse of the determined at least one parameter value;
adjusting the quantization index and repeating the step of tentatively coding the index until the upper and lower probability values for the tentatively coded index match the decoded pair of upper and lower probability values from the arithmetic bitsream; and
outputting the quantization index that resulted in a match.
9. The method of claim 8 , wherein the quantization indices to be decoded represent quantized frequency coefficients of an audio signal, the envelope function modeling the spread of coded quantization indices as a function of frequency, the method including performing a frequency-to-time transformation of the output quantization indices to obtain digitized audio samples of an audio signal.
10. The method of claim 8 , wherein the quantization indices to be decoded represent quantized spatial frequency coefficients of a pixel based image, the envelope function modeling the spread of coded quantization indices as a function of spatial frequency, the method including performing a frequency-to-space transformation of the output quantization indices to obtain digitized pixel data of a pixel based image.
11. The method of claim 8 , wherein said upper and lower probability values are values of a cumulative probability density function.
12. The method of claim 11 , wherein the probability model of the cumulative probability density function is based on a distribution included in the set of: Gaussian probability distribution, a logistic distribution and a Laplace distribution.
13. The method of claim 8 , wherein said at least one envelope parameter include an envelope scale parameter.
14. The method of claim 13 , wherein the envelope model is an Auto-Regressive, AR, model and the envelope parameters include AR coefficients.