A Semantic Image Retrieval Framework Based on Ontology and Naïve Bayesian Inference
Roung Shiunn Wu, Wen Hsien Hsu, Chia Jia Yi · 2012
To deal with large amounts of multimedia content, the MPEG-7 standard was proposed in order to provide a general standard by which to describe various types of multimedia content. However, we required not only a standard to describe multimedia content but also a retrieval framework to search for the requested semantic content. The proposed framework employs ontology and MPEG-7 descriptors to deal with problems arising between syntactic representation and semantic retrieval of images. Instead of building a single ontology for a specific domain, the framework allows for the construction of incrementally multiple ontologies, and shares ontology information not only between the image seekers but also between different domains. Naive Bayesian inference is used to estimate the similarity between query ontology and domain ontology for matching relevant images. The framework provides a relevant feedback mechanism for image seekers to respond to relevant images in the matching process, thus enabling the framework to enhance image annotations for improved image retrieval precision.