Back to results
US 12,762,179 B2Granted
Systems and methods for image content indexing and search
AssigneeOpen Text Corporation
InventorsRavi Ranjan Jha, Dhiman Karmakar
Life of this patent
Estimating the term…
Abstract
A computer-implemented method for content search, comprising accessing an image stored or to be stored in a repository that is searchable using a search index, applying a machine learning model to detect objects and actions in the image and updating the search index to index the image using the objects and actions detected in the image to enable searching of the image by the objects and actions detected in the image.
Claims23
A content search system, comprising:
- a repository for storing content;
- an index database comprising a search index for the repository;
- a processor coupled to the repository and the index database;
- a computer memory storing code executable by the processor for:
- accessing a selected image stored or to be stored in the repository;
- applying a machine learning model to the selected image to detect objects and an action being performed in the selected image, wherein the machine learning model is trained to detect the objects and the action from individual images, wherein the selected image is an individual still image, and wherein applying the machine learning model comprises:
- applying an object detection model trained to infer object classes to the selected image to detect the objects in the selected image and assign object class labels to the selected image indicating the objects detected in the selected image;
- and
- applying an action detection model trained using the object classes to infer classes of actions being performed to the object class labels assigned by the object detection model to the selected image to detect the action being performed in the selected image and assign an action class label indicating the action being performed detected in the selected image;
- and
- updating the search index to index the selected image using the objects and the action being performed detected in the selected image to enable searching of the selected image by the objects, and the action being performed detected in the selected image, wherein updating the search index comprises adding content indexing labels for the selected image to the search index, the content indexing labels comprising the object class labels indicating the objects detected in the selected image and the action class label indicating the action detected in the selected image.
8 more elements
A computer-implemented method for content search, comprising:
- accessing a selected image stored or to be stored in a repository that is searchable using a search index;
- applying a machine learning model to detect objects and an action being performed in the selected image, wherein the machine learning model is trained to detect the objects and the action from individual images, wherein the selected image is an individual still image, and wherein applying the machine learning model comprises:
- applying an object detection model trained to infer object classes to the selected image to detect the objects in the selected image and assign object class labels to the selected image indicating the objects detected in the selected image;
- and
- applying an action detection model trained using the object classes to infer classes of actions being performed to the object class labels assigned by the object detection model to detect the action being performed in the selected image and assign an action class label to the selected image indicating the action being performed detected in the selected image;
- and
- updating the search index to index the selected image using the objects and the action detected in the selected image to enable searching of the selected image by the objects and the action being performed, detected in the selected image, wherein updating the search index comprises adding content indexing labels for the selected image to the search index, the content indexing labels comprising the object class labels indicating the objects detected in the selected image and the action class label indicating the action being performed detected in the selected image.
4 more elements
Dependencies are inferred from the claim text.