Scalable form matching
Disclosed are a method and apparatus for determining a given template of a form used by a filled in instance of that type of form from amongst a great number of form templates (a hundred or more). The given instance is evaluated by a neural network that has been trained by a single example of each template in order to reduce the total number of templates down to a manageable amount. Given a list of closest matching templates, the instance is aligned to each of the closest matching templates. The comparison generates a match score. The form template having the greatest match score is the correct form template. Filtering the instance through a one-shot learning neural network before performing a precise comparison enables the process to scale to any number of template forms.
1 . A method of identifying a form template from a set of templates that a filled-in instance of a form corresponds to comprising:
executing a first operation via a recognition neural network trained on one copy of each of the set of templates that identifies a subset of the set of templates that most closely corresponds to the filled-in instance of the form;
executing a second operation via a registration process that identifies a best match of the subset of the set of templates and the filled-in instance of the form, wherein the second operation is more computationally complex than the first operation; and
sorting the filled-in instance of the form with forms corresponding to the best match of the subset of the set of templates, wherein said executing the first operation further comprises: generating a set of template vectors describing graphical dimensions of each of the set of templates; generating a query vector describing graphical dimensions of the filled-in instance of the form; and scoring the query vector against each of the set of template vectors via the recognition neural network.
2 . The method of claim 1 , wherein the filled-in instance of the form is a member of a set of filled-in forms, the method further comprising:
filtering out blank images within the set of filled-in forms.
3 . The method of claim 2 , further comprising:
orienting the set of filled-in forms right-side up via rotating each of the set of filled-in forms by 0, 90, 180, or 270 degrees, and no other amounts.
4 . The method of claim 1 , wherein the subset of the set of templates comprises a predetermined number of closest matches.
5 . The method of claim 1 , wherein the subset of the set of templates comprises each of the set of templates that exceed a threshold matching score with the filled-in instance of the form.
6 . The method of claim 1 , wherein the recognition neural network is architecturally configured as a facial recognition neural network.
7 . A system comprising:
a memory that stores a set of template forms;
a processor configured to associate a filled-in instance of a form with a specific form template of the set of template forms based on instructions stored in the memory, the instructions when executed cause the processor to:
execute a first operation via a recognition neural network trained on one copy of each of the set of template forms that identifies a subset of the set of template forms that most closely corresponds to the filled-in instance of the form; and
execute a second operation via a registration process that identifies a best match of the subset of the set of template forms and the filled-in instance of the form, wherein the second operation is more computationally complex than the first operation, wherein the memory further includes instructions that cause said execution of the first operation to further comprise: generate a set of template vectors describing graphical dimensions of each of the set of template forms; generate a query vector describing graphical dimensions of the filled-in instance of the form; and score the query vector against each of the set of template vectors via the recognition neural network.
8 . The system of claim 7 , wherein the filled-in instance of the form is a member of a set of filled-in forms, the processor further configured to:
filter out blank images within the set of filled-in forms.
9 . The system of claim 8 , the processor further configured to orient the set of filled-in forms right-side up via rotating each of the set of filled-in forms by 0, 90, 180, or 270 degrees, and no other amounts.
10 . The system of claim 7 , wherein the subset of the set of template forms comprises a predetermined number of closest matches.
11 . The system of claim 7 , wherein the subset of the set of template forms comprises each of the set of template forms that exceed a threshold matching score with the filled-in instance of the form.
12 . The system of claim 7 , wherein the recognition neural network is architecturally configured as a facial recognition neural network.