Survey of Semantic-Based Image-to-Image Retrieval
Danyang Cao, Hongbo Zhou, Huifang Yang · 2024
Image-to-image retrieval has been a focal point of research in both academia and industry over the past few decades, finding applications across various fields. However, with advancements in technology, early methods have revealed limitations in efficiency and the ability to accurately interpret semantic information within images. Consequently, achieving semantic image-to-image retrieval has become a critical challenge in this domain. This paper reviews recent developments in semantic image retrieval technologies by summarizing and discussing the implementations and pros and cons of various approaches. We begin by introducing commonly used datasets and evaluation metrics pertinent to this task. Next, we present representative algorithms for both single-object image retrieval and complex scene image retrieval, categorized by their application scenarios. Finally, we summarize the pressing issues that remain in current image retrieval technologies and analyze potential future research directions.