Scene text recognition in mobile applications by character descriptor and structure configuration.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Extracting text from natural scenes is difficult due to varied text patterns and background noise.
• Scene text recognition is crucial for numerous applications requiring information extraction from images.

Purpose of the Study:

• To propose an effective method for scene text recognition from detected text regions.
• To develop a compatible algorithm for smart mobile devices and demonstrate its effectiveness.

Main Methods:

• Utilized previously proposed algorithms for initial text region detection.
• Designed a discriminative character descriptor by integrating advanced feature detectors.
• Modeled character structure using stroke configuration maps for each character class.

Main Results:

• Developed an Android-based demo system showcasing scene text extraction from nearby objects.
• Evaluated the proposed method on benchmark datasets, demonstrating its effectiveness.
• Achieved performance comparable to state-of-the-art scene text recognition techniques.

Conclusions:

• The proposed scene text recognition method is effective and comparable to existing approaches.
• The algorithm is suitable for mobile applications, facilitating real-time text extraction.
• Further insights for algorithm improvement were gained through the demo system development.