Text placement within images using neural networks
AssigneeAdobe Inc.
InventorsKate Sousa, Zhe Lin, Saeid Motiian, Pramod Srinivasan, Baldo Faieta, Alex Filipkowski
Life of this patent
Estimating the term…
Abstract
Based on a received digital image and text, a neural network trained to identify candidate text placement areas within images may be used to generate a mask for the digital image that includes a candidate text placement area. A bounding box for the digital image may be defined for the text and based on the candidate text placement area, and the text may be superimposed onto the digital image within the bounding box.
Claims20
A computer program product, the computer program product being tangibly embodied in a non-transitory computer-readable storage medium and comprising instructions that, when executed by at least one computing device, are configured to cause the at least one computing device to:
- receive a digital image and an input text;
- generate a heatmap for the digital image using a neural network trained to identify candidate text placement areas within images based on training data that includes training images and training text superimposed on the training images, the heatmap including pixel values assigned to pixels of the digital image, wherein each pixel value corresponds to a probability of inclusion of at least one candidate text placement area, and wherein the heatmap is generated independently of the input text;
- position a template bounding box at a plurality of locations of the digital image to produce a plurality of candidate bounding boxes based on the heatmap using a sliding window technique, wherein the template bounding box is based on a size or a shape of the input text;
- assign a score for each candidate bounding box of the plurality of candidate bounding boxes based on the pixel values of the heatmap located within each of the plurality of candidate bounding boxes;
- select a bounding box for the input text from the plurality of candidate bounding boxes based the score;
- and
- superimpose the input text onto the digital image by placing the input text within the bounding box.
A computer-implemented method, the method comprising:
- receiving a digital image and input text;
- generating a heatmap for the digital image using a neural network trained to identify candidate text placement areas within images, the heatmap including pixel values assigned to pixels of the digital image, wherein the heatmap is generated independently of the input text;
- positioning a template bounding box at a plurality of locations of the digital image to produce a plurality of candidate bounding boxes based on the heatmap using a sliding window technique, wherein the template bounding box is based on a size or a shape of the input text;
- assigning a score for each candidate bounding box of the plurality of candidate bounding boxes based on the pixel values of the heatmap located within each of the plurality of candidate bounding boxes;
- selecting a bounding box for the input text from the plurality of candidate bounding boxes based the score;
- and
- superimposing the input text onto the digital image by placing the input text within the bounding box.
A system comprising:
- at least one memory including instructions;
- and
- at least one processor that is operably coupled to the at least one memory and that is arranged and configured to execute instructions that, when executed, cause the at least one processor to
- receive a digital image and input text;
- process the digital image and input text with a neural network trained using training data that includes training images and training text superimposed on the training images based on text placement heuristics, to thereby obtain a heatmap for the digital image, wherein the heatmap is generated independently of the input text;
- define a bounding box, based on the heatmap;
- and
- superimpose the input text onto the digital image and by placing the text within the bounding box.
Dependencies are inferred from the claim text.