Helsinki Deblur Challenge 2021: Blurred and Sharp Image Pairs for Testing

University of Helsinki researchers created this dataset for testing image deblurring algorithms. It contains photographs of text strings, natural images, and QR codes, each with a blurred and a sharp version, captured with two cameras. The dataset includes 20 steps of different blur levels, with text images accompanied by ground-truth transcription files.

Use Cases

• Benchmarking image deblurring algorithms based on paired sharp and blurred images.
• Testing optical character recognition (OCR) systems on blurred text targets with provided ground-truth transcriptions.
• Evaluating QR code reader robustness against varying levels of defocus blur.
• Comparing algorithm performance across different blur levels, as the data is split into 20 steps.

Strengths

• Each image has a corresponding sharp or blurred counterpart, providing clear ground truth for training or evaluation.
• Text target images include ground-truth transcriptions in separate .txt files.
• The dataset includes multiple target types: text, natural images, QR codes, and technical targets.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

University of Helsinki

Collection Method

Captured using two cameras (one focused, one misfocused) to create paired sharp and blurred images.