A Real-World Approach on the Problem of Chart Recognition Using Classification, Detection and Perspective Correction.

Area of Science:

• Computer Vision
• Data Visualization
• Image Processing

Background:

• Data charts are ubiquitous in media, aiding data comprehension.
• Static chart images often lack accessible underlying data, hindering analysis and redesign.
• Existing chart recognition methods face challenges with real-world image distortions and variations.

Purpose of the Study:

• To develop and evaluate automatic methods for data extraction from chart images.
• To investigate the utility of perspective detection and correction for enhancing chart recognition.
• To adapt state-of-the-art models for robust chart extraction from real-world photographic data.

Main Methods:

• Proposed a classification, detection, and perspective correction pipeline for chart images.
• Utilized computer vision techniques, specifically perspective detection and correction.
• Trained models on real-world photographic data to address distortions and noise.

Main Results:

• Demonstrated that perspective correction effectively clarifies distorted chart images.
• Showcased improved chart recognition accuracy and efficiency with the proposed methods.
• Indicated that adapted chart recognition models are viable for real-world chart extraction.

Conclusions:

• Automatic data extraction from chart images is feasible using computer vision.
• Perspective correction is a valuable technique for handling real-world chart image challenges.
• The proposed methods offer a practical solution for preparing real-world charts for data extraction.