Deep non-relaxation Hash image retrieval method based on point pair similarity

Abstract

The invention discloses a deep non-relaxation Hash image retrieval method based on point pair similarity. The method comprises the following steps: dividing a data set of an image into a training sample set Dtrain and a test sample set Dtest according to a ratio of 5: 1; constructing a deep convolutional network architecture; training the convolutional neural network by using the training sample set and taking the training data set image and the category label thereof as the input of the neural network to obtain and store a model of the deep neural network; and removing a dropout layer of theconvolutional neural network model according to the convolutional neural network model, and adding a symbol function to the output end of the network; inputting the training sample set Dtrain and theDtest into the model to obtain a Hash code Btrain of the training sample set and a Hash code Btest of the test sample set; and taking the hash code of the test image from the test sample to obtain a vector corresponding to the Hamming distance; and sorting each numerical value of the Hamming distance vector in an ascending order to serve as a retrieval result. According to the method, the problemthat a large number of errors are generated in the process of quantizing the binary hash code through the hash function is effectively solved, and the image retrieval accuracy is improved.

Description

technical field [0001] The invention belongs to the field of deep learning and digital image processing, and more specifically relates to a deep non-relaxation hash image retrieval method based on point pair similarity. Background technique [0002] In recent years, with the development of computer software and hardware technology, the dimensions and quantity of data such as images and videos have been increasing. In order to solve the storage and retrieval problems of massive high-dimensional data, there has been a project of projecting high-dimensional data to low-dimensional binary values. Hashing Learning Methods for Spaces. Hash learning method is a machine learning method that projects high-dimensional space data into low-dimensional Hamming space through hash functions or function clusters under the condition of maintaining the similarity between high-dimensional data such as images or videos. Method, this method uses the hash learning method to index the data, impro...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problems existing in the existing hash learning image retrieval method, the present invention provides a deep non-relaxed hash image retrieval method based on point pair similarity. For the problem of the influence of model accuracy, the symbolic function Moved out of the convolutional neural network model, instead, use in the model Norm and soft threshold function to better constrain quasi-hash codes and achieve more accurate retrieval

Images