Gaze Detection and Text Extraction from Images using CNN with DBSCAN-based Filter Design

Gaze detection and text extraction are pivotal technologies in the domain of human-computer interaction and computer vision, enabling applications such as assistive technologies, user interface optimization, and automated content analysis. This paper presents a novel method that identifies the user’s gaze direction and extracts the necessary portion of text, which constitutes the user’s region of interest, using a Convolutional Neural Network (CNN) integrated with a spatial attention mechanism. A Density-Based Spatial Clustering of Applications with Noise (DBSCAN)-based filter is incorporated into the system to remove textual noise from the extracted image, thereby improving the model’s accuracy and reducing computation time. The proposed method demonstrates reduced dependency on the CNN compared to previous approaches, resulting in enhanced real-time operation capability and lower latency. Computational optimization and real-time performance were thoroughly evaluated, showing that our system not only achieves high accuracy in gaze detection and text extraction but also maintains efficient processing speeds suitable for dynamic applications. This advancement offers significant improvements for applications in assistive technology, educational tools, and interactive media, where precise and efficient gaze detection and text extraction are critical.