Back to results
US 8,442,321 B1Granted
Object recognition in images
AssigneeGoogle Inc.
InventorsEdward Y. Chang, Zhiyu Wang, Dingyin Xia
Life of this patent
Estimating the term…
Abstract
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for labeling images. In one aspect, a method includes automatically identifying an object in an image using a deep model-based and data-driven hybrid architecture.
Claims27
A method implemented by data processing apparatus, the method comprising:
- detecting a plurality of edges in an image under different value combinations of one or more model feature dimensions;
- calculating a measurement of similarity distance of each of the detected edges from one or more image patches in an image patch library, each of the image patches having values representing the model feature dimensions;
- creating a first image vector space, each data point in the first image vector space being based on the measurements of similarity distance of detected edges from a specific image patch in the image patch library and representing a minimal similarity distance of the image from the specific image patch;
- dividing the image into a plurality of first sub-images;
- comparing each first sub-image, using values representing one or more data feature dimensions in the first sub-image, with a plurality of second sub-images in a sub-image library to determine which second sub-images are within a threshold of similarity to the first sub-image, each of the second sub-images having values representing the data feature dimensions;
- creating a second image vector space, each data point in the second image vector space representing a number of first sub-images that are within the threshold of similarity to a second sub-image;
- combining the first and second image vector spaces and mapping the combination to an object vector space in a plurality of object vector spaces, each object vector space being labeled;
- and
- associating the image with the label of the mapped object vector space.
6 more elements
A system comprising:
- a computer storage medium having instructions stored thereon;
- and
- data processing apparatus configured to execute the instructions to perform operations comprising:
- detecting a plurality of edges in an image under different value combinations of one or more model feature dimensions;
- calculating a measurement of similarity distance of each of the detected edges from one or more image patches in an image patch library, each of the image patches having values representing the model feature dimensions;
- creating a first image vector space, each data point in the first image vector space being based on the measurements of similarity distance of detected edges from a specific image patch in the image patch library and representing a minimal similarity distance of the image from the specific image patch;
- dividing the image into a plurality of first sub-images;
- comparing each first sub-image, using values representing one or more data feature dimensions in the first sub-image, with a plurality of second sub-images in a sub-image library to determine which second sub-images are within a threshold of similarity to the first sub-image, each of the second sub-images having values representing the data feature dimensions;
- creating a second image vector space, each data point in the second image vector space representing a number of first sub-images that are within the threshold of similarity to a second sub-image;
- combining the first and second image vector spaces and mapping the combination to an object vector space in a plurality of object vector spaces, each object vector space being labeled;
- and
- associating the image with the label of the mapped object vector space.
9 more elements
A computer readable storage medium having instructions stored thereon which, when executed, cause data processing apparatus to perform operations comprising:
- detecting a plurality of edges in an image under different value combinations of one or more model feature dimensions;
- calculating a measurement of similarity distance of each of the detected edges from one or more image patches in an image patch library, each of the image patches having values representing the model feature dimensions;
- creating a first image vector space, each data point in the first image vector space being based on the measurements of similarity distance of detected edges from a specific image patch in the image patch library and representing a minimal similarity distance of the image from the specific image patch;
- dividing the image into a plurality of first sub-images;
- comparing each first sub-image, using values representing one or more data feature dimensions in the first sub-image, with a plurality of second sub-images in a sub-image library to determine which second sub-images are within a threshold of similarity to the first sub-image, each of the second sub-images having values representing the data feature dimensions;
- creating a second image vector space, each data point in the second image vector space representing a number of first sub-images that are within the threshold of similarity to a second sub-image;
- combining the first and second image vector spaces and mapping the combination to an object vector space in a plurality of object vector spaces, each object vector space being labeled;
- and
- associating the image with the label of the mapped object vector space.
6 more elements
Dependencies are inferred from the claim text.