Enhancing Image Annotation Precision Through Tensor-Based Approaches and Gaussian Filtering in Automated Tagging

In the realm of image analysis for both commercial and research purposes, effectively bridging the gap between visual representation and textual description is crucial. The challenge lies in enhancing image tagging quality while eliminating the need for manual annotation. To address this, a Gaussian filter is employed to enhance the low-level visual features of images, thereby mitigating the semantic gap. Images are converted into tensors to group similar features, and a three-level tucker decomposition is applied to optimize contextual matching within these tensor groups. The tensor formation and context estimation in this proposed approach (EIAPTT-GFAT) play a pivotal role in minimizing the semantic gap. This methodology proves particularly beneficial in addressing issues such as the assignment of irrelevant tags or the omission of relevant ones. The algorithm's efficacy is validated through testing on the corel-10 K dataset, showcasing its potential for advancing automatic image tagging with improved precision and reduced semantic disparity.