MM-FOOD: a high-dimensional index structure for efficiently querying content and concept of multimedia data.

The article presents MM-FOOD, a novel high-dimensional index structure designed to efficiently query multimedia data by integrating semantic (conceptual) and low-level content descriptors into a unified framework. MM-FOOD employs dimensionality reduction—specifically a landmark-based multidimensional scaling method called LMDSFastMap—to embed high-dimensional content features into a low-dimensional space, followed by clustering (using fuzzy c-means, k-means, or X-Tree) and an Array Index mechanism to represent clusters in one dimension. Conceptual attributes are encoded as bit strings and combined with cluster identifiers to form hybrid keys for indexing in a fuzzy object-oriented index structure (FOOD-Index), enabling efficient support for content-based, concept-based, combined, and fuzzy queries. The study compares MM-FOOD with an alternative structure, FOOD-X, which uses separate indices for content (X-Tree) and concept (FOOD-Index), demonstrating that MM-FOOD achieves superior query response times and comparable or better retrieval accuracy on a real-world Turkish news video dataset and a large synthetic dataset. The work highlights the benefits of combining semantic and content information in a single index and the effectiveness of LMDSFastMap for dimensionality reduction in multimedia retrieval tasks.