Sketch Based Image Retrieval

Sketch-based image retrieval (SBIR) aims to find images matching hand-drawn sketches, bridging the semantic gap between abstract drawings and realistic photos. Current research heavily utilizes deep learning models, particularly vision transformers and convolutional neural networks, often incorporating techniques like triplet loss, attention mechanisms, and multi-modal prompt learning to improve accuracy and address challenges such as zero-shot retrieval (handling unseen categories) and fine-grained retrieval (distinguishing similar instances). SBIR's significance lies in its potential for intuitive user interfaces in various applications, including medical image analysis, forensic science, and multimedia search, where traditional keyword-based methods are insufficient.