Identifying versions of a form

Abstract

Disclosed are a method and apparatus for identifying versions of a form. In an example, clients of a medical company fill out many forms, and many of these forms have multiple versions. The medical company operates in 10 states, and each state has a different version of a client intake form, as well as of an insurance identification form. In order to automatically extract information from a particular filled out form, it may be helpful to identify a particular form template, as well as the version of the form template, of which the filled out form is an instance. A computer system evaluates images of filled out forms, and identifies various form templates and versions of form templates based on the images.

1 . A method comprising:
 receiving a plurality of images of forms;   initializing a plurality of clusters based on the plurality of images, wherein each cluster of the plurality of clusters includes an exemplar image that is one of the plurality of images;   aligning exemplar images from two clusters of the plurality of clusters based on keypoints of the exemplar images;   computing a form similarity score based on the alignment of the exemplar images;   when the form similarity score is above a predetermined threshold, merging the two clusters into a same cluster of the plurality of clusters; and   identifying form images from one cluster of the plurality of clusters as being versions of a same template form based on the form images being in the one cluster.   
     
     
         2 . The method of  claim 1 , further comprising:
 determining a single template form associated with each of the plurality of clusters via registration of a single image of each of the plurality of clusters of forms against a set template forms; and   assign all forms in each cluster to a particular template form based on said determining.   
     
     
         3 . The method of  claim 1 , wherein the similarity score is based on a sub-comparisons of predetermined regions of each of the plurality of images, wherein the predetermined regions are based on regions of a set of template forms that are most indicative of variations between a plurality of form versions. 
     
     
         4 . The method of  claim 3 , wherein the predetermined regions are based on the keypoints in the plurality of images. 
     
     
         5 . The method of  claim 1 , wherein said computing is based on a threshold of matching vector elements, wherein the vector elements describe each of the plurality of images. 
     
     
         6 . The method of  claim 5 , wherein said computing further includes:
 assigning a class label to each vector element in the respective image.   
     
     
         7 . The method of  claim 5 , wherein said computing is performed using a trained machine learning model that receives vectors and outputs a confidence of similarity score. 
     
     
         8 . The method of  claim 7 , wherein the confidence of similarity score is based on a “k-nearest neighbors” classification. 
     
     
         9 . A system comprising:
 A processor:   A memory including instructions that when executed cause the processor to:
 receive a plurality of images of forms; 
 initialize a plurality of clusters based on the plurality of images, wherein each cluster of the plurality of clusters includes an exemplar image that is one of the plurality of images; 
 align exemplar images from two clusters of the plurality of clusters based on keypoints of the exemplar images; 
 compute a form similarity score based on the alignment of the exemplar images; 
 when the form similarity score is above a predetermined threshold, merge the two clusters into a same cluster of the plurality of clusters; and 
 identify form images from one cluster of the plurality of clusters as being versions of a same template form based on the form images being in the one cluster. 
   
     
     
         10 . The system of  claim 9 , wherein the instructions further comprise:
 determining a single template form associated with each of the plurality of clusters via registration of a single image of each of the plurality of clusters of forms against a set template forms; and   assign all forms in each cluster to a particular template form based on said determining.   
     
     
         11 . The system of  claim 9 , wherein the similarity score is based on a sub-comparisons of predetermined regions of each of the plurality of images, wherein the predetermined regions are based on regions of a set of template forms that are most indicative of variations between a plurality of form versions. 
     
     
         12 . The system of  claim 11 , wherein the predetermined regions are based on the keypoints in the plurality of images. 
     
     
         13 . The system of  claim 9 , wherein said compute instruction is based on a threshold of matching vector elements, wherein the vector elements describe each of the plurality of images. 
     
     
         14 . The system of  claim 13 , wherein said compute instruction further includes:
 assigning a class label to each vector element in the respective image.   
     
     
         15 . The system of  claim 13 , wherein said compute instruction is performed using a trained machine learning model that receives vectors and outputs a confidence of similarity score. 
     
     
         16 . The system of  claim 15 , wherein the confidence of similarity score is based on a “k-nearest neighbors” classification. 
     
     
         17 . One or more non-transitory computer-readable mediums having executable instructions stored thereon that, when executed by one or more processors, perform the operations of:
 receiving a plurality of images of forms;   initializing a plurality of clusters based on the plurality of images, wherein each cluster of the plurality of clusters includes an exemplar image that is one of the plurality of images;   aligning exemplar images from two clusters of the plurality of clusters based on keypoints of the exemplar images;   computing a form similarity score based on the alignment of the exemplar images;   when the form similarity score is above a predetermined threshold, merging the two clusters into a same cluster of the plurality of clusters; and   identifying form images from one cluster of the plurality of clusters as being versions of a same template form based on the form images being in the one cluster.   
     
     
         18 . The one or more non-transitory computer-readable mediums of  claim 17 , further comprising:
 determining a single template form associated with each of the plurality of clusters via registration of a single image of each of the plurality of clusters of forms against a set template forms; and   assign all forms in each cluster to a particular template form based on said determining.   
     
     
         19 . The one or more non-transitory computer-readable mediums of  claim 17 , wherein the similarity score is based on a sub-comparisons of predetermined regions of each of the plurality of images, wherein the predetermined regions are based on regions of a set of template forms that are most indicative of variations between a plurality of form versions. 
     
     
         20 . The one or more non-transitory computer-readable mediums of  claim 19 , wherein the predetermined regions are based on the keypoints in the plurality of images. 
     
     
         21 . The one or more non-transitory computer-readable mediums of  claim 17 , wherein said computing is based on a threshold of matching vector elements, wherein the vector elements describe each of the plurality of images. 
     
     
         22 . The one or more non-transitory computer-readable mediums of  claim 21 , wherein said computing further includes:
 assigning a class label to each vector element in the respective image.   
     
     
         23 . The one or more non-transitory computer-readable mediums of  claim 21 , wherein said computing is performed using a trained machine learning model that receives vectors and outputs a confidence of similarity score. 
     
     
         24 . The one or more non-transitory computer-readable mediums of  claim 23 , wherein the confidence of similarity score is based on a “k-nearest neighbors” classification.