Research On Image Retrieval Based On Hashing And Generative Adversarial Networks

The era of artificial intelligence continues to advance,enabling computers to operate in the same way as humans in many ways,greatly improving the efficiency of doing things.Because of advances in technology,image retrieval methods are also constantly improving,and the demand for image retrieval is increasing.Therefore,how to improve speed and accuracy in large-scale image retrieval tasks has become the most critical task in image retrieval engineering.Aiming at how to improve the accuracy and efficiency of large-scale image retrieval tasks,an image retrieval model combining hash algorithm and generative adversarial networks algorithm is proposed.The model is carried out under unsupervised conditions,that is,the problem of similar vector classification is not completed by using image label information,so that the retrieval precision is accurate and the retrieval rate is efficient.Firstly,the convolutional neural network is used to extract image features.The deeper features of the image can be effectively extracted to make the image more complex.The content is accurately expressed.Secondly,the obtained feature vector is quantized by the hash function to obtain a binary hash code.At the same time,in order to correctly classify the unlabeled image and the hash code of the similar image is more similar,the neighboring structure is constructed according to the classification algorithm.The hash algorithm generates a binary hash code,then,using the binary hash code generated by the generative adversarial networks optimization,the binary noise variable is used as the input of the generator,so that the discriminator judges the generated fake sample or the real obtained binary sample.The generator and the discriminator pass the continuous game.The accurate hash code is output,and finally the similarity matching is performed on the feature vector of the image by using the hamming distance,and finally the image retrieval task is completed.Experiments show that the experiment is suitable for large-scale image retrieval tasks,and experimental verification on the standard image dataset MNIST handwritten image dataset and CIFAR-10 image dataset,compared with the experimental results of the classic unsupervised hash algorithm for image retrieval.The experimental results show that the model greatly improves the retrieval accuracy compared to the comparative retrieval model.