Image Retrieval Method Based on Multi-task Hashing Learning

Abstract

The invention discloses an image retrieval method based on multi-task hash learning. First determine the deep convolutional neural network model; secondly, use the multi-task learning mechanism to design the loss function; then determine the convolutional neural network model training method, combine the loss function, and use the backpropagation method to optimize the model; finally input the image to the convolution Neural network model, and convert the output of the model into a hash code for image retrieval. The convolutional neural network model is composed of a convolutional sub-network and a fully connected layer; the convolutional sub-network is composed of a first convolutional layer, a maximum pooling layer, a second convolutional layer, an average pooling layer, and a third volume The base layer and the spatial pyramid pooling layer are composed sequentially; the fully connected layer is composed of a hidden layer, a hash layer and a classification layer. The model training method has two training methods of combined and separated. The method of the invention can effectively retrieve single-label and multi-label images, and the retrieval performance is better than other deep hashing methods.

Description

technical field [0001] The invention belongs to the technical field of image retrieval and relates to an image retrieval method based on multi-task hash learning. Background technique [0002] Considering the ever-increasing digital image resources on the Internet, using linear search to retrieve information in such a huge image library will cause huge computing and storage overheads. Therefore, in the application of CBIR technology, the "curse of dimensionality" "Problems happen from time to time. In order to solve this problem, in recent years, the approximate nearest neighbor search has gradually become a part of the focus of researchers, and the hash method is a typical representative of it. The goal of the hash method is usually to let the initial image data be calculated by a hash function to obtain a series of fixed-length binary codes, so as to achieve dimensionality reduction for image representation and help reduce storage overhead. In the phase of similarity cal...

AI Technical Summary

Problems solved by technology

[0009] The purpose of the present invention is to provide an image retrieval method based on multi-task hash learning, which solves the problem of the deep hash image retrieval method in the prior art, which leads to insufficient accuracy of retrieval images due to insufficient utilization of sample supervision information. High problem, effectively improve the accuracy of image retrieval

Images