Spectrally adaptive noise filling tool (SANFT) for perceptual transform coding of still and moving images

A decoder apparatus for decoding a picture from a data stream, configured to:

• derive a quantized transform from the data stream, a zero-quantized portion of the quantized transform, and at least one noise parameter providing noise information for a transform in the positions of the zero-quantized portion;
• provide noise synthesis to synthesize noise associated with the quantized transform on the basis of the at least one noise parameter;
• and
• reconstruct the picture using the quantized transform and the synthesized noise, wherein the decoder apparatus is configured to:
• define at least one frequency boundary on the basis of a noise colorness parameter or granularity parameter describing colorness or granularity of noise of the transform in the positions of the zero-quantized portion, the noise colorness parameter or granularity parameter being comprised in the at least one noise parameter;
• determine, from the zero-quantized portion, a higher-frequency zero-value subportion and a lower-frequency zero-value subportion;
• and
• set at zero the noise for the higher-frequency zero-value subportion and/or avoid noise insertion for the higher-frequency zero-value subportion and/or prevent from synthesizing noise for the higher-frequency zero-value subportion.

An encoding method, comprising:

• quantizing a transform of a block of a picture to acquire a quantized transform;
• defining a second block having coefficients of the transform associated to a zero-quantized portion of the quantized transform;
• acquiring at least one noise parameter providing noise information on the transform within the zero-quantized portion;
• signal the quantized transform and the at least one noise parameter, wherein the method comprises:
• dividing the second block into a higher-frequency zero-quantized portion and a lower-frequency zero-quantized portion;
• and
• determining the noise colorness parameter or granularity parameter on the basis of a comparison and/or a relationship and/or a quotient based on:
• a first sum or aggregate value or statistical value or average value associated to the transform within the higher-frequency zero-quantized portion;
• and
• a second sum or aggregate value or statistical value or average value associated to the transform within the second block or within a super block comprising the block.

A decoding method, comprising:

• acquiring a quantized transform from a data stream, a zero-quantized portion of the quantized transform, and at least one noise parameter providing noise information for the zero-quantized portion;
• performing noise synthesis associated with the quantized transform on the basis of the at least one noise parameter;
• and
• reconstructing a picture using the quantized transform and the synthesized noise, wherein the method further comprises:
• defining at least one frequency boundary on the basis of a noise colorness parameter or granularity parameter describing colorness or granularity of noise of a transform in the positions of the zero-quantized portion, the noise colorness parameter or granularity parameter being comprised in the at least one noise parameter;
• determining, from the zero-quantized portion, a higher-frequency zero-value subportion and a lower-frequency zero-value subportion;
• and
• setting at zero the noise for the higher-frequency zero-value subportion and/or avoid noise insertion for the higher-frequency zero-value subportion and/or prevent from synthesizing noise for the higher-frequency zero-value subportion.