Image Retrieval Based on Multi-feature Fusion

In content-based image retrieval, and for this critical issue of image feature fusion, paper proposes a new method to determine the weights for multi-feature fusion. In this paper, color histogram, color correlogram, gray level co-occurrence matrix, Tamura and Hu moments, this five kinds of feature extraction method was adopted. Firstly, use these five features conducted single feature retrieval on the various types of images to determine the precision rate of each feature retrieval and compare their precision rate. Through precision rate to determine the dynamic weight of various features when conducting the feature fusion retrieval in different categories images. The experimental results showed that: according the precision rate of each feature to dynamically regulate the weights, when carrying multi-feature fusion retrieval for different types of image, compared to multi-feature retrieval with fixed weights, precision rate of retrieval has improved significantly.