Turkish Scene Text Recognition Dataset

The Turkish Scene Text Recognition (TS-TR) dataset, addresses the critical gap in non-English text recognition resources, focusing on the unique linguistic features of the Turkish language. This dataset is designed to handle specific challenges such as Turkish diacritics and special characters. It features text captured in real-world settings, showcasing a diverse range of fonts, sizes, orientations, and complex backgrounds from both urban and rural environments. The dataset ensures the creation of robust models capable of recognizing text across various lighting conditions, visual layouts, and environmental contexts, offering significant advantages for applications in AI-driven text recognition solutions.

Key Features:

• Text Variety: The dataset includes text from street signs, billboards, shop names, and more. These samples come from both urban and rural areas, ensuring models can generalize to various real-world environments.

• Linguistic Diversity: It contains Turkish text with special characters and diacritics, making it essential for training models that can recognize complex script forms.

• Visual Complexity: The dataset features text in different orientations, font styles, and sizes. It also includes challenging background conditions, like crowded scenes and varying lighting, ensuring models can handle diverse visual environments.

• Large Scale & Rich Annotations: With thousands of labeled samples, the dataset includes bounding boxes around text regions, transcription data, and metadata. This enables both scene text detection and recognition tasks.

• Real-World Relevance: The dataset is sourced from real-world scenarios, reflecting practical challenges like those faced by autonomous systems, mobile-based text recognition, and document processing.

Applications:

• Scene Text Detection & Recognition: Perfect for training models that extract and recognize Turkish text in real-world imagery, such as street signs and documentation.

• Augmented Reality (AR): This dataset is valuable for developing AR systems that require real-time text recognition in physical environments.

• Autonomous Systems: The dataset helps train AI models for autonomous vehicles or drones, enabling them to read street signs and advertisements in both urban and rural settings.

• Document Digitization: The dataset supports AI-driven document processing tools that convert Turkish text from photos into digital formats.

• Language-Specific OCR: Ideal for building Optical Character Recognition (OCR) systems tailored to Turkish, where English-based OCR tools may struggle with language-specific nuances.