US 7,587,092 B2
Layer-based context quantization with context partitioning
Hua Cai, Kowloon (Hong Kong Special Administrative Region of the People's Republic of China, The); and Jiang Li, Beijing (China)
Assigned to Microsoft Corporation, Redmond, Wash. (US)
Filed on Nov. 25, 2005, as Appl. No. 11/287,054.
Prior Publication US 2007/0122046 A1, May 31, 2007
Int. Cl. G06K 9/36 (2006.01); G06K 9/46 (2006.01)
U.S. Cl. 382—239  [382/238] 16 Claims
OG exemplary drawing
 
1. A method for selecting conditioning states to be used in a context model for coding a source, comprising:
storing in a memory the method for selecting conditioning states to be used in the context model for coding the source;
processing on a processor the method for selecting conditioning states to be used in the context model for coding the source;
recognizing a plurality of potential conditioning states;
partitioning the plurality of potential conditioning states into groups of conditioning states according to a rule, at least one of the groups including a plurality of conditioning states, wherein the potential conditioning states includes one of:
a plurality of original conditioning states representing possible combinations of context events; and
a plurality of quantized conditioning states quantized from the potential conditioning states or a plurality of previously quantized conditioning states;
selecting at least one of the groups of conditioning states to be used in the context model; and
forming a plurality of layers by:
collecting the conditioning states in a layer; and
quantizing at least a portion of the conditioning states in the layer to create a next layer of collective conditioning states, wherein quantizing the conditioning states in the layer to create the next layer includes:
quantizing each of a plurality of pairs of conditioning states in the layer as a collective conditioning state in the next layer;
when there is a remaining conditioning state in the layer after each of the pairs of conditioning states have been quantized, including the remaining conditioning state in the next layer; and
outputting the selected conditioning states to be used in the context model for coding a source.