Study and application of semantic-based image retrieval

Abstract

Image retrieval is increasingly necessary for multi-media resources to improve user experiences and expand retrieval approach. This paper studies into recent process made in semantic-based image retrieval (SBIR), and establishes a semantic-based image retrieval model (SBIRM) for ‘Classic Reading’ Platform. The model chooses three important features from illustrations of books including color, texture and shape to extract feature vector and form a 31-dimentional vector. Vector normalization is performed to eliminate the difference between different features and similarity measurement is used to compare two images. Finally, K-nearest neighbor (KNN) was performed to train the image vectors to decide which book the given image belongs to. The experimental result shows that the precision of this model can be more than 90%.