Abbyy Software Limited

Abbyy Development Inc.

Abbyy Production LLC

Vladimir Rybkin

Konstantin Anisimovich

Ivan Germanovich Zagaynov

Olga Arnoldova Kacher

Vadim Tereshchenko

Sergey Fedorov

Dmitry Vnuchkov

Dmitry Deryagin

Aleksey Zhuravlev

Alexander Shamis

Anisimovich Konstantin Vladimirovich

Davletshin Azat Aydarovich

Sergey Platonov

Azat Aydarovich Davletshin

Zhuravlev Aleskey Alekseevich

Olga Arnoldovna Kacher

A classification engine generates, using a weighted graph, a plurality of sets of confused graphemes based on recognition data for a plurality of document images; receives an input grapheme image associated with a document image comprising a plurality of grapheme images; determines a set of recognition options for the input grapheme image, where the set of recognition options comprises a set of target characters that are similar to the input grapheme image; identifies a neural network trained to recognize a first set of confused graphemes, where the first set of confused graphemes comprises at least a portion of the set of recognition options for the input grapheme image; and determines a grapheme class for the input grapheme image using the identified neural network.

Differential classification using multiple neural networks

The current document is directed to methods and systems that acquire an image containing text with curved text lines to generate a corresponding corrected image in which the text lines are straightened and have a rectilinear organization. The method may include identifying a page sub-image within the text-containing image, generating a text-line-curvature model for the page sub-image that associates inclination angles with pixels in the page sub-image, generating local displacements, using the text-line-curvature model, for pixels in the page sub-image, and transferring pixels from the page sub-image to a corrected page-sub-image using the local displacements to construct a corrected page sub-image in which the text lines are straightened and in which the text characters and symbols have a rectilinear arrangement.

Method and system for preparing text images for optical-character recognition

A classification engine stores a plurality of neural networks in memory, where each neural network is trained to recognize a set of confused graphemes from one or more sets of confused graphemes identified in recognition data for a plurality of document images. The classification engine receives an input grapheme image associated with a document image comprising a plurality of graphemes, determines a set of recognition options for the input grapheme image, wherein the set of recognition options comprises a set of target characters that are similar to the input grapheme image, selects a first neural network from the plurality of neural networks, wherein the first neural network is trained to recognize a first set of confused graphemes, and wherein the first set of graphemes comprises at least a portion of the set of recognition options for the input grapheme image, and determines a grapheme class for the input grapheme image using the selected first neural network.

The current document is directed to methods and systems that straighten in the text lines of text-containing digital images. Initial processing of a text-containing image identifies the outline of a text-containing page. Next, aggregations of symbols, including words and word fragments, are identified within the outlined page image. The centroids and inclination angles of the symbol aggregations are determined, allowing each symbol aggregation to be circumscribed by a closest-fitting rectangle oriented in conformance with the inclination angle determined for the circumscribed symbol aggregation. A model is constructed for the text-line curvature within the text image based on the circumscribed symbol aggregations and is refined using additional information extracted from the text image. The model, essentially an inclination-angle map, allows for assigning local displacements to pixels within the page image which are then used to straighten the text lines in the text image.

The current document is directed to methods and systems that straighten curvature in the text lines of text-containing digital images, including text-containing digital images generated from the two pages of an open book. Initial processing of a text-containing image identifies the outline of a text-containing page. Next, contours are generated to represent each text line. The midpoints and inclination angles of the links or vectors that comprise the contour lines are determined. A model is constructed for the perspective-induced curvature within the text image. In one implementation, the model, essentially an inclination-angle map, allows for assigning local displacements to pixels within the page image which are then used to straighten the text lines in the text image. In another implementation, the model is essentially a pixel-displacement map which is used to straighten the text lines in the text image.

Method and system that efficiently prepares text images for optical-character recognition

An original image of a physical document is received by a client device, a reduced file containing data indicative of a document type is created based on the original image that is substantially smaller in size than the original image. The client device sends a request including the reduced file to a server for information pertaining to a document type for the physical document, receives location information based on the document type from the server for at least one portion of the original image that contains at least one content item for the physical document, and extracts the at least one portion of the original image based on the location information to generate at least one extracted portion of the image. The client device sends a second request including the at least one extracted portion of the image to the server for the at least one content item. Responsive to receiving the at least one content item from the server device, the client device provides the at least one content item for display.

Optimizing data exchange between client device and server

A page orientation component of an image processing device receives an image of a document, transforms the image to a binarized image by performing a binarization operation on the image, and identifies a portion of the binarized image that comprises one or more rows of textual content. The page orientation component identifies a plurality of horizontal runs of white pixels and a plurality of vertical runs of white pixels in the one or more rows of textual content in the portion of the binarized image. The page orientation component generates a first histogram for the plurality of horizontal runs of white pixels, and a second histogram for the plurality of vertical runs of white pixels, and determines an orientation of the one or more rows of textual content in the image based on the first histogram and the second histogram.

Determining the direction of rows of text

Methods for correcting distortions in an image including text, or an image of a page that includes text, are disclosed. The methods include identifying reliable and substantially straight lines from elements in the image. Vanishing points are determined from the lines. Parameters associated with a rectangle are determined. A coordinate conversion is performed.

Straightening out distorted perspective on images

Disclosed is a method of bit-mapped image analysis that comprises a whole image data representation via its component objects. The objects are assigned to different levels of complexity. The objects may be hierarchically connected by spatially-parametrical links. The method comprises preliminarily generating a classifier of image objects consisting of one or more levels differing in complexity; parsing the image into objects; attaching each object to one or more predetermined levels; establishing hierarchical links between objects of different levels; establishing links between objects within the same level; and performing an object feature analysis. Object feature analysis comprises generating and examining a hypothesis about object features and correcting the concerned object's features of the same and other levels in response to results of hypothesis examination. Object feature analysis may also comprise execution of a recursive X-Y cut within the same level.

Multilevel image analysis

Embodiments of the present invention disclose a method, device and system for restoring a motion-blurred image. The method comprises determining parameters for a one-dimensional Optical Transfer Function (OTF) for the motion-blurred image in Fourier space; determining a signal-to-noise ratio for the motion-blurred image in the Fourier space; and correcting for motion blur based on the parameters of the OTF. Determining the parameters comprises calculating a function &#x3a6;(p,q) which is based on the square of the modulus of the Fourier transform |G(p,q)|of the motion-blurred image. The parameters include the absolute value of the one-dimensional OTF, and the phase and sign of the OTF.

Device and system for restoring a motion-blurred image

In one embodiment, a method for correcting distortions in a scanned image of a page is disclosed. The method comprises identifying at least one set of collinear elements in the scanned image; and generating a corrected image based on the scanned image including for at least some of the collinear elements in each set applying a spatial location correction to position all collinear elements in the set on a common horizontal rectilinear base line in the corrected image.

Method and system for straightening out distorted text-lines on images

The present invention discloses a multilevel method of bitmapped image analysis that comprises a whole image data representation via its components&#x2014;objects of different levels of complexity&#x2014;hierarchically connected therebetween by spatially-parametrical links. The said method comprises preliminarily generating a classifier of the objects that possibly may be present in the image consisting of one or more levels differing in complexity; parsing the image into objects; attaching each object to one of predetermined levels; establishing hierarchical links between objects of different levels; establishing links between objects within the same level; and performing an object feature analysis. The objects feature analysis comprises at least generating and examining a hypothesis about object features and correcting the object's features of the same and other levels in response to the hypothesis examination results. The step of object features analysis may also comprise execution of a recursive RX-cut within the same level.

Multilevel bit-mapped image analysis method

Embodiments of the present invention disclose a method and system for restoring a motion-blurred image. The method comprises determining parameters for a one-dimensional Optical Transfer Function (OTF) for the motion-blurred image in Fourier space; determining a signal-to-noise ratio for the motion-blurred image in the Fourier space; and correcting for motion blur based on the parameters of the OTF. Determining the parameters comprises calculating a function &#x3a6;(p,q) which is based on the square of the modulus of the Fourier transform |G(p,q)|of the motion-blurred image. The parameters include the absolute value of the one-dimensional OTF, and the phase and sign of the OTF.

Method and system for restoring a motion-blurred image

The present invention discloses a method of character and text recognition of a bit-mapped graphic file received from an optical scanning device. The method comprises a trainable template cache, a preliminarily trained feature analysis means, and a context analysis means. The present invention discloses the way to use said means for achieving the best results in recognition. The method supposes that the template cache along with the context analysis means are used as the main shape characteristic analyzing means. The feature analysis means along with the context analysis means are used as subsidiary shape characteristic analyzing means and as a training means for the template cache. The method comprises applying the main shape characteristic analyzing means and optionally applying the subsidiary shape characteristic analyzing means if no or not enough reliability of recognition is achieved after the template cache analyzing. The obtained results are analyzed and sent to the template cache for template training.

Growing community of inventors

Moscow, Russia

Vladimir Rybkin