Graded image retrieval method based on deep features of convolutional neural network

Abstract

The invention provides a graded image retrieval method based on deep features of a convolutional neural network. According to the basic principle, the method comprises the steps of firstly, training the convolutional neural network used for feature extracting, and determining network parameters; then using the trained convolutional neural network to extract image features, and obtaining multiple convolutional layer binary system features and one full-joint layer binary system feature; secondly, applying the convolutional layer binary system features to a preliminary screening retrieval stage,conducting multi-feature similarity fusion after further compressing the features, sifting out a candidate image set, and reducing the retrieval range; finally, using the full-joint layer binary system feature to accurately retrieve the candidate image set to obtain a final retrieval result. As is shown by an experiment result based on a public image retrieval dataset, compared with an existing image retrieval method, the representing mode of the images of the graded image retrieval method is more comprehensive, a feature compression method is simpler and more efficient, and the retrieval accuracy is high; meanwhile, by means of a graded retrieval mode, the system calculation amount is dispersed, parallel accelerating is achieved, and the graded image retrieval method has practical value.

Description

technical field [0001] The invention belongs to the field of image processing technology and information retrieval, and relates to a hierarchical image retrieval method realized by extracting deep features using a convolutional neural network in deep learning. Background technique [0002] With the explosive growth of image data, fast and effective retrieval of image data is an important way to manage massive image data. Content-Based Image Retrieval (CBIR) technology should be used under such actual needs. And born. CBIR is an image retrieval method that achieves matching by extracting image content information. Its goal is: given a query image by the user, quickly retrieve images related to the content of the query image from a large-scale database, and follow the The similarity ranking is returned to the user. [0003] The traditional CBIR system implements the image retrieval function by manually extracting the visual features in the image, such as color, texture, shap...

AI Technical Summary

Problems solved by technology

[0007] In order to solve the problem that the traditional CBIR system does not have a deep image representation, there is a "semantic gap" in the extracted features, and the existing CNN-based CBIR system does not fully utilize the network features, the feature compression algorithm is not suitable for two-dimensional features and the algorithm complexity is high. Problem, the present invention proposes a kind of image retrieval method based on CNN depth feature

Images