<p>Fast and accurate image annotation is essential for enabling smart indexing in real-time image processing applications, such as intelligent search engines, smart surveillance, and augmented reality systems. As digital imagery continues to grow at an unprecedented scale, the need for scalable, intelligent, and responsive annotation frameworks becomes increasingly critical. This paper introduces a real-time multi-label image annotation framework that leverages a quantum-inspired optimization strategy to address these challenges. At its core, the system employs a nonparametric Bayesian classifier to manage inter-concept visual similarity and intra-concept diversity—two major hurdles in semantic image understanding. Images are represented as bags of features derived from Otsu-based segmented regions (blobs), enabling robust object-level analysis. To enhance the precision of multi-threshold segmentation, the proposed method integrates the quantum firefly algorithm (QFA), which extends the traditional binary firefly algorithm using principles from quantum computation. QFA improves convergence speed, prevents premature stagnation, and ensures search diversity by simulating quantum particle behavior within the optimization process. Semantic richness is further enhanced through localized labeling, linking frequently occurring textual descriptors in the dataset with their corresponding visual regions. The proposed method achieves low computational complexity and robust global optimization performance, making it particularly well suited for real-time environments. Extensive experiments on the Corel image datasets confirm the framework’s superior performance in terms of label ranking accuracy, convergence efficiency, and resistance to local optima. These results underscore the potential of QFA-based annotation for real-time semantic image retrieval in next-generation intelligent Internet systems.</p>