One-Shot Template Matching for Automatic Document Data Capture

TL;DR

This paper introduces a one-shot template-matching algorithm that automatically extracts data from business documents with minimal manual input, demonstrating high accuracy on invoice datasets.

Contribution

The paper presents a novel one-shot template-matching method that uses engineered features to extract data from documents with the same format, invariant to position and value changes.

Findings

Achieved 86.4% accuracy on invoice dataset

Effective in extracting data with minimal manual annotation

Invariant to positional and value variations

Abstract

In this paper, we propose a novel one-shot template-matching algorithm to automatically capture data from business documents with an aim to minimize manual data entry. Given one annotated document, our algorithm can automatically extract similar data from other documents having the same format. Based on a set of engineered visual and textual features, our method is invariant to changes in position and value. Experiments on a dataset of 595 real invoices demonstrate 86.4% accuracy.