Landmark-Based Document Recognition for Faster Template Search

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing media input devices in self-service kiosks face a bottleneck in recognition speed due to exhaustive document template comparison, which hinders high-throughput performance despite achieving high recognition accuracy.

Innovation Solution

Reorganizing document templates by defining anchor descriptors and clustering them into landmarks, limiting the search space to a small group of templates affiliated with these landmarks, and ordering groups by frequency of occurrence to accelerate recognition.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If exhaustive document template comparison is used, then recognition accuracy is improved, but recognition speed deteriorates

Engineering Contradiction:
Improverecognition accuracyVSAvoidrecognition speed
Core Design Contradiction:
Measurement precisionVSSpeed

Solution Approach 1:

The patent segments the document template comparison process into two distinct phases: (1) a coarse filtering phase that rapidly eliminates non-matching document types using simplified criteria, and (2) a fine matching phase that performs detailed template comparison only on the small subset of candidates that passed filtering. This segmentation resolves the contradiction by maintaining high accuracy through the fine matching phase while achieving high speed through the aggressive filtering in the first phase.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary actions by pre-processing and organizing document templates into hierarchical categories and pre-computing feature descriptors before actual recognition occurs. During runtime, the system first performs preliminary filtering based on coarse features (such as document size, color distribution, or basic structural elements) to eliminate obviously non-matching templates, then proceeds to detailed comparison only for remaining candidates. This preliminary action significantly reduces the comparison space while preserving recognition accuracy.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If exhaustive document template comparison is used, then recognition accuracy is improved, but throughput deteriorates

Engineering Contradiction:
Improverecognition accuracyVSAvoidthroughput
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent segments the template comparison workload into multiple processing stages with decreasing complexity. The first stage performs rapid filtering on all incoming documents using lightweight features, the second stage performs detailed comparison only on filtered candidates, and subsequent stages handle validation. This segmentation enables high throughput by processing the majority of documents through fast paths while maintaining accuracy through thorough processing of relevant candidates.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies partial action by performing complete template comparison only when necessary, rather than on every document. The system uses confidence thresholds and progressive refinement strategies where documents that clearly match or clearly don't match are handled with less computational effort, while only ambiguous cases receive the full exhaustive comparison treatment. This approach maintains accuracy for critical cases while improving overall throughput.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3518191B1Rapid landmark-based media recognition
Publication Date: 2025.10.01 NCR VOYIX CORP
  • EP3518191B1 patent drawingFigure 1
  • EP3518191B1 patent drawingFigure 2
  • EP3518191B1 patent drawingFigure 3

AI summary

Various embodiment herein each include at least one of systems, devices, methods, and software for rapid landmark-based media recognition. One such embodiment, in the form of a method includes receiving a document image comprising pixels and processing the pixels of the document image to identify landmarks present therein. The method then selects a document template classification group based on the landmarks identified within the document image and compares the document image to document templates of the selected document template classification group to classify a document type of the received image. Some embodiments of the method further include forwarding the document image and the document type classification to a document type validation process.