Xerox Corporation

Conduent Business Services, LLC

Safwan R Wshah

Reiner Eschbach

Beilei Xu

Peter Paul

Martin Sidney Maltz

Orhan Bulan

Michael Robert Campanelli

Edward Neil Chapman

Jess Robert Gentner

Phillip J Emmett

Eric Michael Gross

Dennis Lee Venable

Saurabh Prabhat

Jayant Kumar

Gary Walker Skinner

Isaiah Lorenzo Simmons, Jr

Yingbo Zhou

According to exemplary methods of training a convolutional neural network, input images are received into a computerized device having an image processor. The image processor evaluates the input images using first convolutional layers. The number of first convolutional layers is based on a first size for the input images. Each layer of the first convolutional layers receives layer input signals comprising features of the input images and generates layer output signals that include signals from the input images and ones of the layer output signals from previous layers within the first convolutional layers. Responsive to an input image being a second size larger than the first size, additional convolutional layers are added to the convolutional neural network. The number of additional convolutional layers is based on the second size in relation to the first size. The additional convolutional layers are initialized using weights from the first convolutional layers. Feature maps comprising the layer output signals are created.

System and method for expanding and training convolutional neural networks for large size input images

An image processing system for identifying table captions may include a processor and a non-transitory computer-readable storage medium comprising one or more instructions configured to cause the processor to perform image processing of a scanned document to identify boxes of an electronic fillable form that corresponds to the scanned document. The system creates and stores a data structure that includes a box list. The system searches the box list to identify an empty box, uses an adjacency list determine whether the empty box has adjacent boxes, and in response to determining that the empty box has one or more adjacent boxes, creates a table box list that includes the empty box and one or more other boxes from the table to which the empty box belongs. The system uses the adjacency list to identify a row caption and a column caption associated with the empty box.

Image processing system and methods for identifying table captions for an electronic fillable form

A method and system for domain adaptation based on multi-layer fusion in a convolutional neural network architecture for feature extraction and a two-step training and fine-tuning scheme. The architecture concatenates features extracted at different depths of the network to form a fully connected layer before the classification step. First, the network is trained with a large set of images from a source domain as a feature extractor. Second, for each new domain (including the source domain), the classification step is fine-tuned with images collected from the corresponding site. The features from different depths are concatenated with and fine-tuned with weights adjusted for a specific task. The architecture is used for classifying high occupancy vehicle images.

Multi-layer fusion in a convolutional neural network for image classification

A system of associating text items with fillable fields of an electronic fillable form may identify an electronic fillable form comprising one or more fillable fields and one or more text items, and for one or more pairs of fillable fields and text items, determine a confidence value associated with the pair. The system may determine a first configuration for the electronic fillable form, determine an overall confidence value for the first configuration, and identify one or more orphan fields from the determined configuration. The system may determine a second configuration for the electronic form, determines an overall confidence value for the second configuration, and in response to the overall confidence value for the first configuration exceeding the overall confidence value for the second configuration, selects the first configuration, otherwise, selects the second configuration. The system creates an association map for the selected configuration.

Relational database for assigning orphan fillable fields of electronic fillable forms with associated captions

A method of creating a confidence map for an electronic fillable form may include identifying, by an electronic device, one or more fillable fields of an electronic fillable form. The method may include, for one or more of the identified fillable fields, assigning one or more first coordinates of the electronic fillable form to define a graphical zone associated with the fillable field, assigning one or more second coordinates of the electronic fillable form to define a fill zone, assigning one or more third coordinates of the electronic fillable form to define a click zone, and assigning, by the electronic device, one or more fourth coordinates of the electronic fillable form to define a start zone representing a most likely selection area for the fillable field. The method may include creating, a confidence map associated with the electronic fillable form.

Methods and systems of creating a confidence map for fillable forms

A system of identifying one or more fillable fields of an electronic form may include an electronic device, and a computer-readable storage medium that includes one or more programming instructions. The programming instructions are configured to instruct the electronic device to receive an electronic form, identify fillable field candidates of the electronic form, and determine, for each fillable field candidate, whether the fillable field candidate is a fillable field. The system updates metadata associated with the electronic form by applying a sequencing framework to only the fillable fields by obtaining position information for each fillable field that indicates a position of the fillable field on the document, sorting the fillable fields based on the position information to form a sequence of fillable fields, determining a designator to each fillable field that indicates a position of a corresponding fillable field in the sequence, and storing the designator.

Method and system of identifying fillable fields of an electronic form

A method of determining placement of an input indicator on an electronic fillable form may include identifying, by an electronic device, a confidence map associated with an electronic fillable form. The confidence map may include, for one or more fillable fields of the electronic fillable form, a set of coordinates that define an area associated with each of one or more zones. The method may include receiving, by the electronic device, a selection of a portion of the electronic fillable form associated with one or more selection coordinates of the electronic fillable form, identifying a zone to which the selection coordinates correspond, and causing an input indicator to be displayed in a fill zone associated with the identified zone.

Methods and systems of applying a confidence map to a fillable form

A device for registration of content in a filled-out application form is disclosed. The device is configured for scanning at least one portion of the filled-out application form. The device is configured for extracting filled-out content from the scanned form. The geometrical features of the master form are retrieved. The master form includes one or more anchor fields. Each anchor field has one or more anchor zones and at least one anchor segment. At least one anchor segment has global adjustment parameters and geometrical features. The extracted filled-out content is related to the retrieved geometrical features of a master form to create a new geometrical representation of the extracted filled-out content of the scanned application form. The new representation of the filled-out content based on the global adjustment parameters for the at least one anchor segment is globally adjusted. The globally adjusted filled-out content is locally adjusted.

Global registration of filled-out content in an application form

An image is received that contains handwritten characters as answers to an inquiry. An automatic character recognition process is performed on the image to generate initially recognized characters from the handwritten characters. The initially recognized characters of an incorrect answer to a question of the inquiry are analyzed to automatically identify alternative recognized characters of the incorrect answer. Then, it can be determined whether one or more of the alternative recognized characters, when substituted in place of the initially recognized characters of the incorrect answer, change the incorrect answer to a correct answer to the question. If the alternative recognized characters substituted in place of the initially recognized characters in the correct answer exceed a minimum character recognition confidence value, the incorrect answer is changed to the correct answer to modify the initially scored answers to modified scored answers, and such are output.

Confirming automatically recognized handwritten answers

A method, non-transitory computer readable medium, and apparatus for classifying machine printed text and handwritten text in an input are disclosed. For example, the method defines a perspective for an auto-encoder, receives the input for the auto-encoder, wherein the input comprises a document comprising the machine printed text and the handwritten text, performs an encoding on the input using an auto-encoder to generate a classifier, applies the classifier on the input and generates an output that separates the machine printed text and the handwritten text in the input based on the classifier in accordance with the perspective.

Method and apparatus for classifying machine printed text and handwritten text

A device for registration of content in a filled-out application form is disclosed. The device is configured for scanning at least one portion of the filled-out application form. The device is configured for extracting filled-out content from the scanned form. The geometrical features of the master form are retrieved. The master form includes one or more anchor fields. Each anchor field has one or more anchor zones and at least one anchor segment. At least one anchor segment has global adjustment parameters and geometrical features. The extracted filled-out content is related to the retrieved geometrical features of a master form to create a new geometrical representation of the extracted filled-out content of the scanned application form. The new representation of the filled-out content based on the global adjustment parameters for the at least one anchor segment is globally adjusted. The globally adjusted filled-out content based on the geometrical features for the anchor segments is locally adjusted.

Methods and devices for form-independent registration of filled-out content

Systems and methods scan a previously printed sheet using a scanner to create an original scan, and evaluate the original scan to identify marks printed on the printed sheet. For example, any marks, such as the straight and curved marks located in separation areas (areas separating data areas of the printed sheet) of the printed sheet can be evaluated using a specialized image processor. Then, these systems and methods alter the pixels of such marks to add security features (e.g., microtext, microprinting, etc.) to the marks that identify the time, place, and user of the scanner, and this process produces an altered scan. Rather than outputting the original scan, these systems and methods instead substitute the altered scan in place of the original scan and, therefore, output the altered scan from the image processor in place of the original scan.

Growing community of inventors

Webster, NY, United States of America

Safwan R Wshah