Prediction-Based Image Processing

A method of compressing a block of multiple pixels each having a respective property value, said method comprising the steps of:

• hierarchically defining a subblock in said block based on property values of a subset of said multiple pixels or of a set of hierarchically lower subblocks in said block;
• determining a property value for said subblock based on at least a portion of said property values of said subset of said multiple pixels or of said set of hierarchically lower subblocks;
• providing a prediction of said determined property value by:
• a) determining a first variance representation of property value(s) of at least one neighboring pixel or subblock in a first prediction direction in said block and property value(s) of at least one neighboring pixel or subblock in a second prediction direction in said block;
• b) determining said prediction based on a weighted combination of at least one property value of said at least one neighboring pixel or subblock in said first prediction direction and at least one property value of said at least one neighboring pixel or subblock in said second prediction direction if said first variance representation is smaller than a first variance threshold;
• c) selecting a prediction direction of said first and second prediction directions if said first variance representation is not smaller than said first variance threshold;
• d) determining said prediction to be based on at least one property value of said at least one neighboring pixels or subblocks in said selected prediction direction if said first variance representation is not smaller than said first variance threshold;
• e) providing a first guiding bit associated with said selected prediction direction if said first variance representation is not smaller than said first variance threshold;
• calculating a prediction error based on said determined property value and said provided prediction;
• and
• determining an encoded representation of said prediction error, wherein a compressed representation of said block comprises said encoded representation and said optional first guiding bit.

A method of decompressing a compressed pixel block, said method comprising the steps of:

• determining, for a pixel subblock of said block, a prediction error based on an encoded error representation associated with said subblock and comprised in said compressed pixel block;
• providing a prediction of a property value of said subblock by:
• a) determining a first variance representation of property value(s) of at least one neighboring pixel or subblock in a first prediction direction in said block and property value(s) of at least one neighboring pixel or subblock in a second prediction direction in said block;
• b) calculating said prediction based on a weighted combination of at least one property value of said at least one neighboring pixel or subblock in said first prediction direction and at least one property value of said at least one neighboring pixel or subblock in said second prediction direction if said first variance representation is smaller than a first variance threshold;
• c) selecting a prediction direction of said first prediction direction and said second prediction direction based on a first guiding bit associated with said subblock and comprised in said compressed pixel block if said first variance representation is not smaller than said first variance threshold;
• d) calculating said prediction to be based on at least one property value of said at least one neighboring pixels or subblocks in said selected prediction direction if said first variance representation is not smaller than said first variance threshold;
• calculating a representation of said property value of said subblock based on said prediction error and said prediction;
• and
• assigning said representation of said property values to multiple pixels in said block organized into said subblock.

A block compressor for compressing a block of multiple pixels each having a respective property value, said block compressor comprises:

• a subblock definer for hierarchically defining a subblock in said block based on property values of a subset of said multiple pixels or of a set of hierarchically lower subblocks in said block;
• a value determiner for determining a property value for said subblock based on at least a portion of said property values of said subset of said multiple pixels or of said set of hierarchically lower subblocks;
• a prediction provider for providing a prediction of said determined property value, said prediction provider comprising:
• a variance determiner fur determining a first variance representation of property value(s) of at least one neighboring pixel or subblock in a first prediction direction in said block and property value(s) of at least one neighboring pixel or subblock in a second prediction direction in said block;
• a first prediction determiner adapted to operate if said first variance representation is smaller than a first variance threshold and arranged for determining said prediction based on a weighted combination of at least one property value of said at least one neighboring pixel or subblock in said first prediction direction and at least one property value of said at least one neighboring pixel or subblock in said second prediction direction;
• a direction selector adapted to operate if said first variance representation is not smaller than said first variance threshold and arranged for selecting a prediction direction of said first and second prediction directions;
• a second prediction determiner adapted to operate if said first variance representation is not smaller than said first variance threshold and arranged for determining said prediction to be based on at least one property value of said at least one neighboring pixels or subblocks in said selected prediction direction;
• a guiding bit provider adapted to operate if said first variance representation is not smaller than said first variance threshold and arranged for providing a first guiding bit associated with said selected prediction direction;
• an error calculator for calculating a prediction error based on said determined property value and said provided prediction;
• and
• a representation determiner for determining an encoded representation of said prediction error, wherein a compressed representation of said block comprises said encoded representation and said optional first guiding bit.

A block decompressor for decompressing a compressed pixel block, said block decompressor comprises:

• an error determiner for determining, for a pixel subblock of said block, a prediction error based on an encoded error representation associated with said subblock and comprised in said compressed pixel block;
• a prediction provider for providing a prediction of a property value of said subblock and comprising:
• a variance determiner for determining a first variance representation of property value(s) of at least one neighboring pixel or subblock in a first prediction direction in said block and property value(s) of at least one neighboring pixel or subblock in a second prediction direction in said block;
• a first prediction calculator adapted to operate if said first variance representation is smaller than a first variance threshold and arranged for calculating said prediction based on a weighted combination of at least one property value of said at least one neighboring pixel or subblock in said first prediction direction and at least one property value of said at least one neighboring pixel or subblock in said second prediction direction;
• a direction selector adapted to operate if said first variance representation is not smaller than said first variance threshold and arranged for selecting a prediction direction of said first prediction direction and said second prediction direction based on a first guiding bit associated with said subblock and comprised in said compressed pixel block;
• a second prediction calculator adapted to operate if said first variance representation is not smaller than said first variance threshold and arranged for calculating said prediction to be based on at least one property value of said at least one neighboring pixels or subblocks in said selected prediction direction;
• a representation calculator for calculating a representation of said property value of said subblock based on said prediction error and said prediction;
• and
• a value assigner for assigning said representation of said property values to multiple pixels in said block organized into said subblock.