A novel approach using the local sketch and its variations for image retrieval in education

In the field of education, retrieving student information from classrooms is extremely important. While textual data such as dates and class schedules are commonly used for information retrieval tasks, the use of camera footage in classrooms is also widespread. However, image retrieval, particularly using sketches, is a new and complex technology. In this paper, we develop a sketch-based image retrieval system to extract information from classroom cameras. The final results allow for precise retrieval previously unattainable, enabling users to make increasingly detailed queries and incorporate attributes such as color and contextual hints from the sketches. To achieve this, we introduce a new framework that effectively integrates sketch images using pre-trained CLIP models, eliminating the need for detailed sketch descriptions. Lastly, our system extends to include sketch-based image retrieval applications, domain attribute transformation, and detailed image generation, offering solutions for various real-world scenarios.