A Comprehensive Survey of Cross-Modal Retrieval: Breaking Through Modal Barriers

In the digital era of rapidly expanding multimedia content, effectively retrieving and utilizing information from multimodal data has become crucial. Cross-Modal Retrieval (CMR) overcomes the limitations of single-modality data by enabling mutual retrieval across different modalities, offering richer and more flexible information access for users. CMR bridges the semantic gap between modalities, facilitating the connection and understanding of diverse data types, which plays a vital role in multimodal research. In the survey, we offer a comprehensive examination of cross-modal retrieval, starting with the fundamental concepts and significance of the field. It delves into various methods including supervised, unsupervised, and semi-supervised, categorized by the level of supervision. Additionally, the review explored the emerging trend of cross-modal retrieval under the low-quality data conditions. Finally, we discuss long-standing challenges such as scalability and efficiency in managing large-scale multimedia datasets and explore potential future research challenges in realistic application scenarios.