Image Semantic Extraction Using Latent Semantic Indexing on Image Retrieval Automatic-Annotation

摘要

This research proposed a new method Latent Semantic Indexing to overcome semantic problem on image retrieval based on automatic image annotation. Statistical machine translation used to automatically annotates the image. This approach considers image annotation as the translation of image regions to words, similar to the translation of text from one language to another. The "lexicon" for the translation is learned from large annotated image collections, which consist of images that are associated with text. Images are segmented into regions with grid segmentation. A pre-specified feature vector represents each region. The regions then clustered into a finite set of blobs. The correspondences between the blobs and the words are learned using Expectation Maximization algorithm. These correspondences are used to predict words associated with whole images (auto-annotation). The auto-annotation performance is evaluated using Normalized Score (ENS) algorithm. The experimental results show that the average precision of clause queries achieved best result than word queries, 0.544 and 0.251 for clause queries and word queries respectively. The proposed method of latent semantic indexing succeeds to exploit semantic value of automatic-annotation-based image retrieval.