Research On Image Retrieval With Hash Codes Based On Deep Learning

With the rapid development of Internet and multimedia technology,image data has been growing exponentially.Quickly searching for a specific image from a massive image database has become a hot issue in the field o f image retrieval.Deep learning has strong learning ability in image feature learning,and hash algorithm shows unique advantages in information retrieval,therefore,image retrieval based on deep learning has become the main research method of image retrieval system.The traditional complex deep learning network model has increased the difficulty of learning the hash function,in this paper,we propose a method based on the improved AlexNet neural network model by improving the loss function and adding the full connection layer.In addition,in order to solve the problem of too many parameters in the full connection layer of the network model,this paper also proposes an image retrieval method based on the NIN network model to avoid training overfitting phenomenon and further improve the system image retrieval effect.Specific work can be summarized as:We propose an image retrieval method based on improved AlexNet network model.The main idea of this paper is to add a full connection layer after the AlexNet network model to learn the hash function and binary hash code of image.Using the idea of transfer learning will be trained network model in ImageNet dataset to fine-tune the target data set,so that the trained network model on the existing large data set is applied to the new data set.Meanwhile,an effective loss function is used to maximize the differentiability of the output space by encoding the supervised information from the input image pairs,and imposing regularization on the real-valued outputs to approximate the desired discrete values.We propose an image retrieval method based on NIN model.Traditional generalized linear models have poor applicability to nonlinear separable data,therefore,this paper uses the MLP convolution layer instead of the traditional convolution layer to improve the ability of extracting abstract features and generalization.For the improved AlexNet network model,there are many parameters of the full connection layer,and it is easy to have the problem of fitting.The global mean pooling method is proposed to replace the whole connection layer in traditional CNN.Global mean pooling can greatly reduce the parameters in the network and avoid fitting.We extensively evaluate the retrieval performance on two large-scale datasets CIFAR-10 and NUS-WIDE.And the evaluation shows that our method gives a better performance than traditional hashing learning methods in image retrieval.