Fast image retrieval method based on deep learning

Abstract

An embodiment of the present invention discloses a fast image retrieval method based on deep learning. The method comprises: randomly generating two images from an image database as input of a network, and taking one as a query image and the other as a sample image, wherein each picture comprises a corresponding category tag; constructing a convolutional neural network, wherein the network comprises three sets of convolution pooling layers and two sets of fully connected layers; randomly combining the training sample sets into data pairs to be trained according to the convolutional network, obtaining the corresponding hash code, and calculating the Euclidean distance between two training sample sets; calculating the error function of the output value of the convolutional network, trainingthe convolutional neural network, and updating the network parameters by using the back propagation algorithm and the stochastic gradient descent method; and after obtaining the binary coding of the training data sets, sorting the training data sets according to the Euclidean distances from near to far, and outputting retrieval results in sequence. According to the method disclosed by the embodiment of the present invention, the problems of a slow retrieval speed, a large memory space, and an inaccurate retrieval result in the prior art can be solved, and the space-time efficiency of image retrieval is greatly improved.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence, in particular to a fast image retrieval method based on deep learning. Background technique [0002] In recent years, due to different user needs, thousands of pictures are uploaded on the Internet every day. Due to the increase in the number of pictures, it becomes extremely difficult for users to find the pictures they want. For example, semantic-based image retrieval CBIR (Content Based Image Retrieval) retrieves images similar to a given query image in a database. The "similarity" mentioned here may be semantic similarity, or it may be similar in appearance. It is assumed that all images in the dataset and the image being queried are represented by corresponding feature descriptors as corresponding high-dimensional feature vectors. The easiest way to find the query image is to sort the images in the dataset according to the distance between the query image and the images in ...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to overcome the deficiencies of the prior art. The present invention provides a fast image retrieval method, which can solve the problems of slow retrieval speed, large memory space occupation, and inaccurate retrieval results in the prior art, and greatly improves the efficiency of image retrieval. Space-time efficiency

Images