Apparatus and methods for extracting data from lineless tables using Delaunay triangulation and excess edge removal

A method, comprising:

• receiving, at a processor, a scanned image including lineless formatted data;
• detecting, via optical character recognition (OCR), a plurality of boundary boxes associated with the lineless formatted data;
• generating a graph based on the plurality of boundary boxes and using Delaunay triangulation, the graph including a plurality of edges, each edge from the plurality of edges representing a relationship between exactly two boundary boxes from the plurality of boundary boxes;
• performing, via the processor, a first refinement of the graph to eliminate redundant representations of relationships, to produce a first refined graph;
• performing, via the processor, a second refinement of the first refined graph to remove a longest edge from each triangular region from a plurality of triangular regions produced by the Delaunay triangulation, to produce a second refined graph;
• and
• predicting row data and column data for the lineless formatted data using a neural network and based on the second refined graph.

A method, comprising:

• obtaining, at a processor, a portable document format (PDF) file including formatted data;
• converting the PDF file, via the processor, to an image file;
• performing optical character recognition (OCR) on the image file, via the processor, to produce a scanned file;
• generating, via the processor, a plurality of character-level bounding boxes for the formatted data based on the scanned file;
• generating, via the processor, a plurality of word-level bounding boxes for the formatted data based on the PDF file, the plurality of character-level bounding boxes and the plurality of word-level bounding boxes forming a synthetic dataset, and
• training a neural network using the synthetic dataset.