Detecting the bounds of borderless tables in fixed-format structured documents using machine learning

A system for detecting borderless tables in a fixed-format structured document, the system comprising a processor that is adapted to:

• receive a document comprising a plurality of text lines and a borderless table having a table boundary that defines a perimeter of the borderless table;
• generate a sorted list of the plurality of text lines within the document, wherein the list is sorted based upon a vertical position and a horizontal position of each line in the list
• determine a first confidence value associated with a first one of the text lines that does not form part of the borderless table;
• determine a second confidence value associated with a second one of the text lines, wherein the first and second text lines are associated with different confidence values;
• make a first determination that (a) the first confidence value is less than a threshold and (b) the first text line intersects the table boundary of the borderless table, and in response to making the first determination, enlarge the table boundary of the borderless table to include the first text line within the borderless table;
• and
• make a second determination that the second confidence value is greater than the threshold, and in response to making the second determination, enlarge the table boundary of the borderless table to include the second text line within the borderless table,
• wherein the borderless table includes at least one line having a confidence value greater than the threshold, and at least one line having a confidence value less than the threshold.

A computer program product including one or more non-transitory computer readable media encoded with instructions that, when executed by one or more processors, cause a borderless table detection process to be carried out, the process comprising:

• generating a sorted list of a plurality of text lines within a document, wherein the list is sorted based upon a vertical position and a horizontal position of each text line in the list;
• determining a first respective probability that a first text line belongs to a borderless table that comprises first and second borderless table objects, wherein the first text line comprises a plurality of words, and wherein the first respective probability is based on positions of the words included in the first text line;
• determining a second respective probability that a second text line belongs to the borderless table, wherein the first and second text lines have different probabilities of belonging to the borderless table;
• making a first determination that the first respective probability is greater than a threshold, and in response to making the first determination, designating the first text line as belonging to the borderless table, associating the first text line with the first borderless table object, and updating a boundary of the borderless table;
• making a second determination that (a) the second respective probability is less than the threshold, and (b) the second text line intersects the second borderless table object, and in response to making the second determination, designating the second text line as belonging to the borderless table and associating the second text line with the second borderless table object;
• and
• in response to making the second determination, updating the boundary of the borderless table to reflect the second text line as belonging to the borderless table, wherein the borderless table includes at least one line having a probability greater than the threshold, and at least one line have a probability less than the threshold.