Zero communication block partitioning
OwnerAlban Lefebvre, Axel Loewe +2inventors
InventorsAlban Lefebvre, Axel Loewe, Mariappan S. Nadar, Jun Liu
Life of this patent
Estimating the term…
Abstract
A computer-implemented method for calculating a multi-dimensional wavelet transform in an image processing system comprising a plurality of computation units includes receiving multi-dimensional image data. An overlap value corresponding to a number of non-zero filter coefficients associated with the multi-dimensional wavelet transform is identified. Then the multi-dimensional image data is divided into a plurality of multi-dimensional arrays, wherein the multi-dimensional arrays overlap in each dimension by a number of pixels equal to the overlap value. A multi-dimensional wavelet transform is calculated for each multi-dimensional array, in parallel, across the plurality of computation units.
Claims18
A computer-implemented method for calculating a multi-dimensional wavelet transform in an image processing system comprising a plurality of computation units, the method comprising:
- receiving multi-dimensional image data;
- identifying an overlap value corresponding to a number of non-zero filter coefficients associated with the multi-dimensional wavelet transform;
- dividing the multi-dimensional image data into a plurality of multi-dimensional arrays, wherein the multi-dimensional arrays overlap in each dimension by a number of pixels equal to the overlap value;
- and
- calculating a multi-dimensional wavelet transform for each multi-dimensional array, in parallel, across the plurality of computation units;
- receiving a multi-dimensional wavelet transform result from each of the plurality of computation units, wherein each respective multi-dimensional wavelet transform result excludes values corresponding to overlapping pixels included in the plurality of multi-dimensional arrays;
- and
- aggregating the multi-dimensional wavelet transform results in a memory included in the image processing system.
A method for calculating a transform of a multi-dimensional signal in a system comprising a plurality of computation units, the method comprising:
- dividing, in a first memory associated with a computer, data corresponding to the multi-dimensional signal into a plurality of multi-dimensional arrays, wherein the multi-dimensional arrays overlap in each dimension;
- assigning each multi-dimensional array to one of the plurality of computation units;
- and
- for each computation unit, performing a transform calculation process comprising:
- copying an assigned multi-dimensional array from the first memory to a second memory, wherein the second memory is only accessible to the respective computation unit,
- calculating the transform along each dimension of the assigned multi-dimensional array in the second memory,
- identifying a calculated result of the transform along a last dimension,
- identifying overlap values in the calculated result corresponding to an overlap between the assigned multi-dimensional array and one or more of the plurality of multi-dimensional arrays, and
- copying all values from the calculated result with the exception of the overlap values to the first memory.
An image processing system for calculating a multi-dimensional wavelet transform, the system comprising:
- a receiver configured to receive a multi-dimensional image from an imaging device;
- a first processor comprising a plurality of processor cores;
- and
- a second processor configured to call a software function operable to cause the first processor to:
- divide the multi-dimensional image data into a plurality of overlapping multi-dimensional arrays,
- calculate a multi-dimensional wavelet transform for each multi-dimensional array, in parallel, across the plurality of processor cores,
- receive a multi-dimensional wavelet transform result from each of the plurality of computation units, wherein each respective multi-dimensional wavelet transform result excludes values corresponding to overlapping pixels included in the plurality of multi-dimensional arrays, and
- aggregate the multi-dimensional wavelet transform results in a memory included in the image processing system.
Dependencies are inferred from the claim text.