System for document layout analysis

Claim 2

A document layout analysis method for determining document structure data from input data including the content and characteristics of regions of at least one page forming the document, the method including the steps of: identifying particular types of sections within the page, wherein the step of identifying sections within the page further comprises the steps of identifying rows and columns within a table, Identifying headers and footers on a page, and classifying the headers and footers based upon a horizontal location; identifying captions, wherein the step of identifying captions further comprises identifying an association between a non-text region of the page and the caption identified; and determining boundaries of at least one column on the page, wherein the step of determining boundaries of at least one column on the page further comprises the steps of identification of text regions, grouping of text regions on the page into sections, grouping of the text regions within a section into the at least one column, determining a column width for the at least one column on the page, and determining, using the column width, the number of columns in each section.

Claim 3

The method of claim 1, wherein the input data represents at least two different pages of a document and said pages are of different sizes and where the method further includes the step of resizing at least one element of the page of the document so that all pages of the document are of a common size.

Claim 4

The method of claim 1, wherein the step of identifying captions further comprises: training a neural net with a predefined set of features; employing a trained neural net to generate an algorithm based upon the predefined set of features; and applying the algorithm generated by the neural net to the input data to identify text regions determined to be captions.

Claim 5

The method of claim 3, wherein the predefined set of features includes a subset of features selected from the set including: centering relationship; projection overlap; italic; bold; font style difference; font size; font size difference; distance to end; distance to beginning; intervening text; text position; text blocking; aspect ratio of text; alignment; and caption location.

Claim 6

The method of claim 3, wherein the step of applying the neural net to the input data includes the step of prefiltering the text regions so as to only apply the neural net to likely caption candidates, where said prefiltering step eliminates text regions that are caption candidates based upon a relative size of the text region that is a caption candidate when compared to the non-text region; and the distance between the caption candidate and the non-text region.

Claim 7

The method of claim 5, wherein the prefiltering step includes: comparing an area of a caption candidate with an area of the non-text region to produce an aspect ratio indicative of the area of the caption candidate relative to the non-text area; and eliminating the caption candidate from further processing by the neural net whenever the aspect ratio exceeds a predefined threshold.

Claim 8

A document analysis method comprising the steps of: digitizing, using an image input terminal, a document to produce rasterized data representing the document; processing the rasterized data representing the document with a character recognition engine, wherein the character recognition engine operates to identify characters and characteristics for each of a plurality of words identified within the rasterized data, and where output of the character recognition engine includes descriptions of characters, fonts, and locations on a page of the document; and determining document structure data from output of the character recognition engine including the content and characteristics of regions of at least one page forming the document, said step of determining document structure data including the steps of identifying sections within the page, wherein the step of identifying sections within the page further comprises the steps of identifying rows and columns within a table, Identifying headers and footers on a page, and classifying the headers and footers based upon a horizontal location, identifying captions, wherein the step of identifying captions further comprises the steps of training a neural net with a predefined set of features, employing a trained neural net to generate an algorithm based upon the predefined set of features, and applying the algorithm generated by the neural net to the input data to identify text regions determined to be captions; and determining boundaries of at least one column on the page, wherein the step of determining boundaries further comprises the steps of identification of text regions, grouping of text regions on the page into sections, grouping of the text regions within a section into the at least one column, determining a column width for the at least one column on the page, and determining, using the column width, the number of columns in each section.

Claim 9

A document analysis method comprising the steps of: digitizing, using an image input terminal, a document to produce rasterized data representing the document; processing the rasterized data representing the document with a character recognition engine, wherein the character recognition engine operates to identify characters and characteristics for each of a plurality of words identified within the rasterized data, and where output of the character recognition engine includes descriptions of characters, fonts, and locations on a page of the document; and determining document structure data from output of the character recognition engine including the content and characteristics of regions of at least one page forming the document, said step of determining document structure data including the steps of identifying sections within the page, wherein the step of identifying sections within the page further comprises the steps of identifying rows and columns within a table, Identifying headers and footers on a page, and classifying the headers and footers based upon a horizontal location, identifying captions, wherein the step of identifying captions further comprises the steps of training a neural net with a predefined set of features, employing a trained neural net to generate an algorithm based upon the predefined set of features, and applying the algorithm generated by the neural net to the input data to identify text regions determined to be captions; and determining boundaries of at least one column on the page, wherein the step of determining boundaries further comprises the steps of identification of text regions, grouping of text regions on the page into sections, grouping of the text regions within a section into the at least one column, determining a column width for the at least one column on the page, and determining, using the column width, the number of columns in each section.

Claim 10

The method of claim 7, wherein the step of identifying captions further comprises: training a neural net with a predefined set of features; employing a trained neural net to generate an algorithm based upon the predefined set of features; and applying the algorithm generated by the neural net to the input data to identify text regions determined to be captions.

Claim 11

The method of claim 8, wherein the step of applying the neural net to the input data includes the step of prefiltering the text regions so as to only apply the neural net to likely caption candidates, where said prefiltering step eliminates text regions that are caption candidates based upon a relative size of the text region that is a caption candidate when compared to the non-text region; and the distance between the caption candidate and the non-text region.