Back to results
US 12,676,019 B2Granted
Adaptive filter generator, identifying lines and table morphology
AssigneeTungsten Automation Corporation
InventorDavid Edward Beery
Life of this patent
Estimating the term…
Abstract
For table structure recognition in document processing, a methodology can use 2-click human interaction to define bounding box of the table and convert OCR data into n-dim positioning for column-wise and row-wise assignment. An adaptive filter can identify and assign column-like morphologies in tables. An adaptive cycle based column/row assignment can select best methodology for each table. Column overlap and merge detection can use post-processing for complex table structures. Data storage and feedback mechanisms can be used for adaptive template predictions.
Claims19
A system comprising:
- a data store configured to store computer executable instructions;
- and
- a hardware processor in communication with the data store, the hardware processor, when executing the computer executable instructions, configured to:
- receive a collection of training document images,
- wherein at least some of the collection of training document images corresponds to real documents supplied by end users such that training occurs persistently using the real documents,
- wherein the system provides results to the end users concurrently with the persistent training;
- establish a filter for application to the training document images by:
- providing an array with an odd number of index entries;
- assigning a sign to a center value for the array;
- assigning the opposite sign to two peripheral values for the array;
- and
- providing array orientation for evaluating vertical and horizontal morphological structure in a document image;
- provide at least one hyperparameter for the filter to allow for feedback and optimization of a structure of the filter over time;
- convolve the filter with data in at least one of the training document images to create a processed image;
- and
- evaluate the processed image to identify table structure.
13 more elements
A system comprising:
- a data store configured to store computer executable instructions;
- and
- a computing device in communication with the data store, the computing device, when executing the computer executable instructions, configured to:
- receive a collection of training document images,
- wherein at least some of the collection of training document images corresponds to real documents supplied by end users;
- and
- use the collection to persistently train a convolution filter having index values and having hyperparameters corresponding to features of filter structure, the hyperparameters comprising at least two of the following:
- a filter size corresponding to a number of index values in a one-dimensional array;
- a center expansion comprising an operator for enlarging a center portion beyond a single index value;
- a multiplier configured to adjust a magnitude of the filter response;
- a ratio characterizing response difference between center and peripheral values in the one-dimensional array;
- and
- a threshold comprising a value below which the system ignores the filter response,
- wherein the system provides results to the end users concurrently with the persistent training.
11 more elements
Dependencies are inferred from the claim text.