Image Retrieval Algorithm based on Multi-feature Fusion

Shijie He

Abstract: With the development of science and technology, more and more image information appears in all aspects of life. Image is a common unstructured data. In order to obtain more abundant information of image, this paper proposes an image retrieval algorithm based on multi-feature fusion. Through dimensionality reduction and weight calculation, different image features are fused to improve the accuracy and accuracy of image retrieval. Firstly, the image is preprocessed and the image features are extracted using three basic models, GLCM, SIFT and CNN. Then, the dimensionality of the image features is reduced using principal component analysis. Finally, the multi-scale attention mechanism is used to fuse the image features. Experiments show that the average retrieval accuracy of the proposed method is 93.74% and the accuracy is 87.84%. It is significantly higher than the single feature search results.

Keywords: Image retrieval; Feature fusion; Principal component analysis; Multi-scale attention mechanism