Universal-denoiser context-modeling component and context-modeling method
AssigneeHewlett-Packard Development Company, L.P.
InventorsIgnacio Ramirez, Erik Ordentlich, Gadiel Seroussi, Marcelo Weinberger
Life of this patent
Estimating the term…
Abstract
In various embodiments of the present invention, a context-based denoiser is applied to each noisy-image symbol embedded within a context to determine a replacement symbol for the noisy-signal symbol. The context-based denoiser includes a context-modeling component that efficiently generates context classes and symbol-prediction classes, assigns individual contexts to context classes and symbol-prediction classes, collects symbol-occurrence statistics related to the generated context classes and symbol-prediction classes, and, optionally, generates noisy-symbol predictions.
Claims19
A context-modeler component of an electronic denoiser, the context-modeler component implemented in hardware, as computer instructions stored in an electromagnetic memory and/or mass-storage device that control operation of the denoiser, or as a combination of hardware and stored computer instructions, the context-modeler component comprising:
- a map-receiving component, within the electronic denoiser, that receives an input noisy-image map and stores the input noisy-image map within an electromagnetic memory, and that optionally receives an input prefiltered map and an input binary mask corresponding to the noisy image and stores the prefiltered map and the input binary mask within the electromagnetic memory;
- context-class-related data that includes one or more of
- wing gradients computed for each symbol embedded within a context observed in one of the input maps,
- an activity level computed for each symbol embedded within a context observed in one of the input maps,
- a texture bitmap computed for each symbol embedded within a context observed in one of the input maps,
- and a gradient angle computed for each symbol embedded within a context observed in one of the input maps;
- and
- a map-generating component, within the electronic denoiser, that generates, using context-class-related data, a metamap that represents assignments of input-map symbols, by the context-modeler component, to symbol-prediction classes and context classes and stores the meta map within an electromagnetic memory, and that optionally generates a predicted-noisy-symbol map and stores the predicted-noisy-symbol map within an electromagnetic memory.
A method for assigning, by an electronic denoiser, a context class and a symbol-prediction class to each of a number of noisy symbols considered in a traversal of an electromagnetically-encoded noise-affected symbol map, the method comprising:
- for each symbol considered in the traversal of a noise-affected symbol map,
- generating, by the electronic denoiser, context-class-related data for the symbol by considering the symbol and a number of neighboring symbols K that together comprise a context for the symbol and storing the context-class-related data into an electromagnetic memory within the electronic denoiser;
- quantizing and/or truncating, by the electronic denoiser, the generated context-class-related data to produce a first set of quantized-and/or-truncated context-class-related data;
- grouping, by the electronic denoiser, the first set of quantized-and/or-truncated context-class-related data together into a data unit to form a symbol-prediction-class descriptor that describes the symbol-prediction class to which the symbol is assigned;
- additionally truncating, by the electronic denoiser, each quantized-and/or-truncated context-class-related data of the first set of quantized-and/or-truncated context-class-related data to produce a second set of truncated, quantized context-class-related data, preserving the most significant bits of each quantized-and/or-truncated context-class-related data;
- and
- grouping, by the electronic denoiser, the second set of truncated, quantized context-class-related data together into a data unit to form a context-class descriptor that describes the context class to which the symbol is assigned and storing the context-class descriptor into an electromagnetic memory within the electronic denoiser.
Dependencies are inferred from the claim text.