Text identification in layered digital content
AssigneeAdobe Inc.
InventorsPunit Singh, Shawn Alan Gaither, Leonard Rosenthol, Jayant Vaibhav Srivastava, Ashutosh Mehra
Life of this patent
Estimating the term…
Abstract
Techniques for text identification in layered digital content are described. In an implementation, an item of digital content is received including a plurality of layers. A text layer is identified in the item of digital content from the plurality of layers. A text channel image is generated by isolating the text layer from the plurality of layers. A text identification is then generated based on the text channel image using a page decomposition model, the page decomposition model trained using machine learning.
Claims20
A method comprising:
- receiving, by a processing device, an item of digital content including a plurality of layers;
- identifying, by the processing device, a text layer in the item of digital content from the plurality of layers;
- generating, by the processing device, a text channel image by isolating the text layer from an image layer of the plurality of layers;
- identifying, by the processing device, a text element in the text channel image and an additional text element in the image layer of the plurality of layers using a machine learning model, the text element in the text channel image overlapping the image layer;
- identifying, by the processing device, one or more types of text corresponding to the text element and the additional text element;
- and
- generating, by the processing device, an annotated item of digital content that includes one or more labels identifying the one or more types of the text by combining the text element and the additional text element and eliminating duplicate identified text from the text element and the additional text element.
A system comprising:
- a memory component;
- and
- a processing device coupled to the memory component, the processing device configured to perform operations including:
- receiving an item of digital content including a text layer and an image layer;
- identifying the text layer in the item of digital content;
- generating a text channel image by separating the text layer from the image layer;
- identifying a text element in the text channel image and an additional text element in the image layer using a page decomposition model trained using machine learning, the text element in the text channel image overlapping the image layer;
- identifying, by the processing device, one or more types of text corresponding to the text element and the additional text element;
- and
- generating an annotated item of digital content that includes one or more labels identifying the one or more types of the text by combining the text element and the additional text element and eliminating duplicate identified text between the text element and the additional text element.
A non-transitory computer-readable medium storing executable instructions, which when executed by a processing device, cause the processing device to perform operations comprising:
- receiving an item of digital content including a text channel and at least one image channel;
- identifying the text channel;
- generating a text channel image by isolating the text channel from the at least one image channel;
- identifying a text element in the text channel image and an additional text element in the at least one image channel using a machine learning model, the text element in the text channel image overlapping the at least one image channel;
- identifying, by the processing device, one or more types of text corresponding to the text element and the additional text element;
- and
- generating an annotated item of digital content that includes one or more labels identifying the one or more types of the text by combining the text element and the additional text element and eliminating duplicate identified text between the text element in the text channel image and the additional text element in the at least one image channel.
Dependencies are inferred from the claim text.