Back to results
US 9,996,890 B1Granted
Detection of items
AssigneeSynapse Technology Corporation
InventorsIan Cinnamon, Bruno Brasil Ferrari Faviero, Simanta Gautam
Life of this patent
Estimating the term…
Abstract
According to an aspect, a system comprises at least one processor, a memory, and a non-transitory computer-readable storage medium storing instructions. The stored instructions are executable to cause the at least one processor to: receive a digital image that represents an object scanned by a detection device, determine a region of the digital image that is likely to contain an item, transform the region of the digital image to an embedding, classify, based on the embedding, the region as containing a known class of known item, and responsive to classifying the region as containing the known class of item: generate a graphical representation based on the known class of item.
Claims20
A method comprising:
- receiving a digital image that represents an object scanned by a detection device;
- resizing the digital image to produce a resized digital image;
- normalizing the resized digital image to produce a normalized digital image;
- determining, by a region proposal neural network, a region of the normalized digital image that is likely to contain an item, wherein determining the region of the digital image likely to contain an item comprises:
- identifying a set of anchor boxes;
- passing each anchor box over each pixel position of the normalized digital image to determine a region proposal score for each combination of pixel position and each anchor box;
- and
- selecting a set of regions corresponding to respective combinations of pixel positions and anchor boxes having highest region proposal scores;
- for each given region in the set of selected regions, transforming, with a classification neural network, the given region of the digital image to an embedding, wherein the embedding comprises a representation of the region in a vector space,
- wherein the vector space is defined by features that were determined, by the classification neural network and based on training image data of a set of known items;
- classifying, with the classification neural network, based on the embedding, the region as containing a known class of item from the set of known items;
- and
- responsive to classifying the region as containing the known class of item:
- generating a graphical representation based on the known class of item.
11 more elements
A system comprising:
- at least one processor;
- a memory;
- and
- a non-transitory computer-readable storage medium storing instructions that are executable to cause the at least one processor to:
- receive a digital image that represents an object scanned by a detection device;
- resize the digital image to produce a resized digital image;
- normalize the resized digital image to produce a normalized digital image;
- determine, by a region proposal neural network, a region of the normalized digital image that is likely to contain an item, wherein to determine the region of the digital image likely to contain an item, the at least one processor to:
- identify a set of anchor boxes;
- pass each anchor box over each pixel position of the normalized digital image to determine a region proposal score for each combination of pixel position and each anchor box;
- and
- select a set of regions corresponding to respective combinations of pixel positions and anchor boxes having highest region proposal scores;
- for each given region in the set of selected regions, transform, with a classification neural network, the given region of the digital image to an embedding, wherein the embedding comprises a representation of the region in a vector space,
- wherein the vector space is defined by features that were determined, by the classification neural network and based on training image data of a set of known items;
- classify, with the classification neural network, based on the embedding, the region as containing a known class of item from the set of known items;
- and
- responsive to classifying the region as containing the known class of item:
- generate a graphical representation based on the known class of item.
15 more elements
A non-transitory computer-readable storage medium storing instructions thereon that are executable to cause at least one processor to:
- receive a digital image that represents an object scanned by a detection device;
- resize the digital image to produce a resized digital image;
- normalize the resized digital image to produce a normalized digital image;
- determine, by a region proposal neural network, a region of the normalized digital image that is likely to contain an item, wherein to determine the region of the digital image likely to contain an item, the at least one processor to:
- identify a set of anchor boxes;
- pass each anchor box over each pixel position of the normalized digital image to determine a region proposal score for each combination of pixel position and each anchor box;
- and
- select a set of regions corresponding to respective combinations of pixel positions and anchor boxes having highest region proposal scores;
- for each given region in the set of selected regions, transform, with a classification neural network, the given region of the digital image to an embedding, wherein the embedding comprises a representation of the region in a vector space,
- wherein the vector space is defined by features that were determined, by the classification neural network and based on training image data of a set of known items;
- classify, with the classification neural network, based on the embedding, the region as containing a known class of item from the set of known items;
- and
- responsive to classifying the region as containing the known class of item:
- generate a graphical representation based on the known class of item.
11 more elements
Dependencies are inferred from the claim text.