Vector-based image processing
AssigneeTelefonaktiebolaget LM Ericsson (publ)
InventorJacob Ström
Life of this patent
Estimating the term…
Abstract
A block of image elements is compressed by determining a parameter representative of a size of a bounding section in vector space encompassing feature vectors associated with the image elements. This size parameter is used to provide a deterministic, pseudo-random pattern of multiple representation vectors encompassed by the bounding section. A vector among multiple representation vectors is selected as representation of the feature vector of an image element. An identifier associated with selected vector is assigned to the image element and included in the compressed block which also comprises representations of the size parameter.
Claims20
A method of compressing a block of image elements, each image element having an associated feature vector, said method comprising the steps of:
- determining at least one parameter representative of a size of a bounding section in a feature vector space encompassing at least a portion of said feature vectors;
- utilizing said at least one parameter as initial seed for providing a deterministic, pseudo-random pattern of N multiple representation vectors encompassed by said bounding section in said feature vector space;
- selecting, for at least one image element and based on a feature vector associated with said image element, a representation vector from said N multiple representations vectors as a representation of said feature vector;
- and
- assigning, to said at least one image element, a vector identifier associated with said selected representation vector.
A method of decoding a feature vector representing an image feature of an image element from a compressed block, said method comprising the steps of:
- providing, based on said compressed block, at least one parameter representative of a size of a bounding section in a feature vector space;
- utilizing said at least one parameter as initial seed for providing a deterministic, pseudo-random pattern of N multiple representation vectors encompassed by said bounding section in said feature vector space;
- and
- selecting, based on a vector identifier associated with said image element and included in said compressed block, a vector from said N multiple representation vectors as a decoded representation of said feature vector.
A compressor for compressing a block of image elements, each image element having an associated feature vector, said compressor comprising:
- a parameter determiner adapted to determine at least one parameter representative of a size of a bounding section in a feature vector space encompassing at least a portion of said feature vectors;
- a pattern provider adapted to utilize said at least one parameter as initial seed to provide a deterministic, pseudo-random pattern of N multiple representation vectors encompassed by said bounding section in said feature vector space;
- a vector selector adapted to select, for at least one image element and based on a feature vector associated with said image element, a representation vector from said N multiple representations vectors as a representation of said feature vector;
- and
- an identifier assigner adapted to assign, to said at least one image element, a vector identifier associated with said selected representation vector, wherein the parameter determiner, pattern provider, vector selector, and identifier assigner are implemented by a computer.
A decoder for decoding a feature vector representing an image feature of an image element from a compressed block, said decoder comprising:
- a parameter provider adapted to provide, based on said compressed block, at least one parameter representative of a size of a bounding section in a feature vector space;
- a pattern provider adapted to utilize said at least one parameter as initial seed to provide a deterministic, pseudo-random pattern of N multiple representation vectors encompassed by said bounding section in said feature vector space;
- and
- a vector selector adapted to select, based on a vector identifier associated with said image element and included in said compressed block, a vector from said N multiple representation vectors as a decoded representation of said feature vector;
- wherein the parameter provider, pattern provider, and vector selector are implemented by a computer.
Dependencies are inferred from the claim text.