Object recognition

Patent metadata

Field	Value
Publication number	US-10380461-B1
Application number	US-201715789789-A
Country	US
Kind code	B1
Filing date	Oct 20, 2017
Priority date	Dec 16, 2015
Publication date	Aug 13, 2019
Grant date	Aug 13, 2019

Abstract

Approaches introduce a pre-processing and post-processing framework to a neural network-based approach to identify items represented in an image. For example, a classifier that is trained on several categories can be provided. An image that includes a representation of an item of interest is obtained. Rotated versions of the image are generated and each of a subset of the rotated images is analyzed to determine a probability that a respective image includes an instance of a particular category. The probabilities can be used to determine a probability distribution of output category data, and the data can be analyzed to select an image of the rotated versions of the image. Thereafter, a categorization tree can then be utilized, whereby for the item of interest represented the image, the category of the item can be determined. The determined category can be provided to an item retrieval algorithm to determine primary content for the item of interest. This information also can be used to determine recommendations, advertising, or other supplemental content, within a specific category, to be displayed with the primary content.

First claim

What is claimed is:

1. A computer-implemented method, comprising:
determining a group of product categories for which historical interaction data is available;
recognizing related product categories of the group of product categories based at least in part on the historical interaction data and at least one shared feature of products in the group of product categories;
determining, based at least in part on the related product categories, an initial set of product categories, the initial set of product categories including a set of higher-level product categories and product subcategories of a respective higher-level product category;
generating a categorization tree that includes a plurality of product categories and links between those product categories, the categorization tree including at least the set of higher-level product categories and the product subcategories, the product subcategories being at least one level lower than the higher-level product categories, wherein products in each product category for the initial set of product categories are compared to a product subcategory threshold, the respective product subcategories including products having a respective categorization vector above the product subcategory threshold being included in the categorization tree and the respective product subcategories including products having a respective categorization vector below the product subcategory threshold being excluded from the categorization tree; and
storing the categorization tree for use in determining related content to a query image, the categorization tree including a plurality of categories and links between those categories.

2. The computer-implemented method of claim 1 , further comprising:
obtaining an indication of an object of interest represented in an image; and
analyzing the image to determine a plurality of categorization values.

3. The computer-implemented method of claim 2 , further comprising:
obtaining a data set including category information for each of a plurality of images, respective category information relating to a category for an item represented in a corresponding image of the plurality of images; and
training a classifier using the category information and corresponding image, the classifier configured to determine the plurality of categorization values associated with the object of interest.

4. The computer-implemented method of claim 2 , wherein analyzing the image further includes:
generating a distribution of probability; and
determining the plurality of categorization values based at least in part on the distribution of probability.

5. The computer-implemented method of claim 2 , further comprising:
identifying, based at least on part on the plurality of categorization values and the categorization tree, a product category from the plurality of product categories that includes classification information to identify items related to the object of interest; and
using the classification information to determine related items associated with the object of interest, the related items corresponding to a subset of the items.

6. The computer-implemented method of claim 1 , further comprising:
obtaining user data representing historical user interaction with respect to items in the plurality of categories; and
identifying a subset of the initial set of categories based at least in part on the user data.

7. The computer-implemented method of claim 1 , further comprising:
determining a subset of higher-level categories that is relevant to a specified category; and
using the subset of the higher-level categories to provide classification information for use in determining related content.

8. The computer-implemented method of claim 1 , further comprising:
recognizing related categories based at least in part on metadata included with a plurality of images, each image of the plurality of images including a representation of an item, the metadata including labeling information that identifies a product category of at least a subset of the items represented in the image.

9. A system, comprising:
at least one processor; and
memory including instructions that, when executed by the at least one processor, enable the system to:
determine a group of product categories for which historical interaction data is available;
recognize related product categories of the group of product categories based at least in part on the historical interaction data and at least one shared feature of products in the group of product categories;
determine, based at least in part on the related product categories, an initial set of product categories, the initial set of product categories including a set of higher-level product categories and product subcategories of a respective higher-level product category;
generate a categorization tree that includes a plurality of product categories and links between those product categories, the categorization tree including at least the set of higher-level product categories and the product subcategories, the product subcategories being at least one level lower than the higher-level product categories, wherein products in each product category for the initial set of product categories are compared to a product subcategory threshold, the respective product subcategories including products having a respective categorization vector above the product subcategory threshold being included in the categorization tree and the respective product subcategories including products having a respective categorization vector below the product subcategory threshold being excluded from the categorization tree; and
store the categorization tree for use in determining related content to a query image, the categorization tree including a plurality of categories and links between those categories.

10. The system of claim 9 , wherein the instructions when executed further enable the system to:
obtain an indication of an object of interest represented in an image; and
analyze the image to determine a plurality of categorization values.

11. The system of claim 10 , wherein the instructions when executed further enable the system to:
identify, based at least on part on the plurality of categorization values and the categorization tree, a product category from the plurality of product categories that includes classification information to identify items related to the object of interest; and
use the classification information to determine related items associated with the object of interest, the related items corresponding to a subset of the items.

12. The system of claim 10 , wherein the instructions when executed further enable the system to:
obtain a data set including category information for each of a plurality of images, respective category information relating to a category for an item represented in a corresponding image of the plurality of images; and
train a classifier using the category information and corresponding image, the classifier configured to determine the plurality of categorization values for associated with the object of interest.

13. The system of claim 10 , wherein the instructions when executed further enable the system to:
generate a distribution of probability; and
determine the plurality of categorization values based at least in part on the distribution of probability.

14. The system of claim 9 , wherein the instructions when executed further enable the system to:
obtain user data representing historical user interaction with respect to items in the plurality of categories; and
identify a subset of the initial set of categories based at l

Assignees

• A9 Com Inc

Inventors

• Ravichandran Avinash Aghoram
• Benitez Matias Omar Gregorio
• Bhotika Rahul
• Helmer Scott Daniel
• Jain Anshul Kumar
• Jia Junxiong
• Nambiar Rakesh Madhavan
• Rybakov Oleg

Classifications

• G06V20/70 Primary

  Labelling scene content, e.g. deriving syntactic or semantic representations · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06F18/24

  Classification techniques · CPC title

• G06F18/24323

  Tree-organised classifiers · CPC title

• G06F18/22

  Matching criteria, e.g. proximity measures · CPC title