SEDIQA: Sound Emitting Document Image Quality Assessment in a Reading Aid for the Visually Impaired

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Assistive Technology

Background:

• Optical Character Recognition (OCR) reading aids offer independence for visually impaired people (VIPs).
• Current reading aids require users to visually target text and capture high-quality images, posing a challenge for VIPs.
• Existing no-reference image quality assessors (NR-IQAs) have not been specifically validated for document images in this context.

Purpose of the Study:

• To develop a novel sound-emitting document image quality assessment metric (SEDIQA) for VIPs.
• To automatically capture the best image for OCR accuracy, overcoming visual targeting limitations.
• To analyze OCR performance degradation and validate SEDIQA against established NR-IQAs.

Main Methods:

• Development of SEDIQA, a no-reference image quality assessor (NR-IQA) that provides auditory feedback.
• Systematic testing of OCR performance under various image degradations.
• Validation of SEDIQA against established NR-IQAs using document images.
• Implementation of a document image enhancement technique.

Main Results:

• SEDIQA consistently selects images with the highest OCR accuracy.
• Significant contributors to OCR accuracy reduction were identified.
• SEDIQA demonstrated effective performance on document images compared to other NR-IQAs.
• The enhancement technique improved OCR accuracy by an average of 22% (up to 68%).

Conclusions:

• SEDIQA provides an effective auditory feedback mechanism for VIPs to capture optimal images for OCR.
• The developed system significantly enhances OCR accuracy, improving accessibility for visually impaired users.
• This work offers valuable insights into image quality assessment for document images and assistive reading technologies.