Document segmentation for optical character recognition
AssigneeInternational Business Machines Corporation
InventorsUdi Barzelay, Ophir Azulai, Inbar Shapira
Life of this patent
Estimating the term…
Abstract
An approach to identifying text within an image may be presented. The approach can receive an image. The approach can classify an image on a pixel-by-pixel basis whether the pixel is text. The approach can generate bounding boxes around groups of pixels that are classified as text. The approach can mask sections of an image that where pixels are not classified as text. The approach may be used as a pre-processing technique for optical character recognition in documents, scanned images, or still images.
Claims14
A computer-implemented method for detecting text within an image, the computer-implemented method comprising:
- training, by a processor, a model to detect text within the image, with a plurality of synthesized noisy images containing text, wherein the model comprises a Unet convolutional neural network architecture;
- receiving, by the processor, an input image;
- generating, by the processor, a pixel probability estimate for each pixel in the input image, based at least in part on the Unet convolutional neural network architecture, wherein the probability estimate is the probability a pixel is part of a segmented text;
- generating, by the processor, a segmentation map based at least in part on the per pixel probability estimate for each pixel in the input image;
- generating, by the processor, one or more bounding boxes around the detected text, based at least in part on the segmentation map;
- and
- masking, by the processor, one or more sections of the input image outside of the bounding boxes.
A computer program product for detecting text within an image, the computer program product comprising:
- one or more non-transitory computer readable storage media and program instructions stored on the one or more non-transitory computer readable storage media, the program instructions comprising program instructions to:
- train a model to detect text within the image, with a plurality of synthesized noisy images containing text, wherein the model comprises a Unet convolutional neural network architecture;
- receive an input image;
- generate a pixel probability estimate for each pixel in the input image, based at least in part on the Unet convolutional neural network architecture, wherein the probability estimate is the probability a pixel is part of a segmented text;
- generate a segmentation map based at least in part on the per pixel probability estimate for each pixel in the input image;
- generate one or more bounding boxes around the detected text, based at least in part on the segmentation map;
- and
- mask one or more sections of the input image outside of the bounding boxes.
A computer system for detecting text within an image, the computer system comprising:
- one or more computer processors;
- one or more computer readable storage media;
- and
- program instructions stored on the one or more computer readable storage media for execution by at least one of the one or more processors, the program instructions comprising:
- train a model to detect text within the image, with a plurality of synthesized noisy images containing text, wherein the model comprises a Unet convolutional neural network architecture;
- receive an input image;
- generate a pixel probability estimate for each pixel in the input image, based at least in part on the Unet convolutional neural network architecture, wherein the probability estimate is the probability a pixel is part of a segmented text generate a segmentation map based at least in part on the per pixel probability estimate for each pixel in the input image;
- generate one or more bounding boxes around the detected text, based at least in part on the segmentation map;
- and
- mask one or more sections of the input image outside of the bounding boxes.
Dependencies are inferred from the claim text.