Guided offset correction for loop restoration in video coding

An apparatus for reducing error in a reconstructed frame having pixels, the apparatus comprising:

• a memory;
• and
• a processor configured to execute instructions stored in the memory to:
• classify the pixels into available offset classes based on a classification scheme, wherein the classification scheme includes multiple classifications associated with respective pixel characteristics, and for a classification of the multiple classifications:
• the classification has a respective plurality of classification classes;
• and
• each of the plurality of classification classes of the classification is defined by respective ranges of values of a pixel characteristic associated with the classification, wherein the available offset classes into which the pixels may be classified are determined as respective combinations of classification classes of the multiple classifications, and
• wherein the instruction to classify the pixels comprises instructions to:
• assign a pixel of the pixels to a respective classification class of at least two classifications of the multiple classifications based on values of the pixel and the respective ranges of values defining the plurality of classification classes of each of the at least two classifications;
• and
• assign the pixel to a single offset class of the available offset classes based on a combination of the classification classes of the at least two classifications to which the pixel is assigned;
• for each offset class of those of the available offset classes that include pixels after classifying the pixels:
• determine an offset value for the offset class;
• apply the offset value for the offset class to each pixel of the offset class resulting in offset-adjusted pixels of the offset class;
• determine, for the offset class, an error reduction in using the offset value for the offset class as compared to omitting the offset value for the offset class, the error reduction based on the pixels of the offset class in the reconstructed frame, the offset-adjusted pixels of the offset class, and co-located source pixels in a source frame decoded to generate the reconstructed frame;
• and
• select, for reducing error in the reconstructed frame, a subset of those of the available offset classes that include pixels after classifying the pixels, the subset having highest values for the error reductions.

A method of reducing error in a reconstructed frame comprising pixels, the method comprising:

• decoding classification parameters of a classification scheme and offset values for the reconstructed frame, from an encoded bitstream, the classification parameters and the offset values encoded by:
• classifying pixels of a reconstructed frame generated by an encoder decoding a source frame into available offset classes based on the classification scheme, wherein the classification scheme includes multiple classifications associated with respective pixel characteristics, and for a classification of the multiple classifications:
• the classification has a respective plurality of classification classes;
• and
• each of the plurality of classification classes of the classification is defined by respective ranges of values of a pixel characteristic associated with the classification, wherein the available offset classes into which the pixels may be classified are determined as respective combinations of classification classes of the multiple classifications, and
• wherein classifying the pixels comprises:
• assigning a pixel of the pixels to a respective classification class of at least two classifications of the multiple classifications based on values of the pixel and the respective ranges of values defining the plurality of classification classes of each of the at least two classifications;
• and
• assigning the pixel to a single offset class of the available offset classes based on a combination of the classification classes of the at least two classifications to which the pixel is assigned;
• for each offset class of those of the available offset classes that include pixels after the classifying:
• determining an offset value for the offset class;
• applying the offset value for the offset class to each pixel of the offset class resulting in offset-adjusted pixels of the offset class;
• and
• determining, for the offset class, an error reduction in using the offset value for the offset class as compared to omitting the offset value for the offset class, the error reduction based on the pixels of the offset class in the reconstructed frame generated by the encoder, the offset-adjusted pixels of the offset class, and co-located source pixels in the source frame;
• selecting, for encoding the source frame, a subset of those of the available offset classes that include pixels after the classifying based on the error reductions;
• and
• encoding, in associated with the source frame, the classification parameters and the offset values, the offset values including at least the offset values for the offset classes of the subset;
• classifying the pixels, based on the classification parameters, into those of the available offset classes selected by the encoder;
• and
• applying a respective offset value for an offset class to each pixel of the offset class within the reconstructed frame to generate an offset-adjusted frame.