Image Retrieval Method Research Based On Deep Learning And Feature Fusion

With the development of network information,the multimedia information such as images and videos has been growing rapidly.For these massive images with rich visual information,how to retrieve the images that the user needed quickly in the largescale image database becomes an important issue in a multimedia information retrieval field of research.The deep convolutional neural network has a strong ability of nonlinear representation,it can understand the image of the deeper level of semantic information,and it is widely used in image recognition,object detection,image segmentation and other fields.In this paper,we summarize the current research progress of image retrieval technology and abroad based on the problem of image retrieval,extract the image features by using the deep learning technology,and improve on the existing contentbased image retrieval methods.The contents of this work can be divided into two parts:(1)Hashing Image Retrieval Based on Multi-layer Feature Map FusionIn most image retrieval tasks,the high-level features extracted from the network may lose a lot of local detail information,which can lead to inaccurate search results.To solve this problem,the local feature fusion extracted from the network is used to fuse features of different layers in the convolutional neural network.It also contains the high-level abstract semantic information extracted from the network and the lowlevel local detail information extracted from the network.At the same time,in order to speed up the retrieval speed,the approximate nearest neighbor search strategy is used.It is an efficient search strategy in the area of large-scale image search.The experiment result shows,compared with the current mainstream approaches,the hashing image retrieval method based on deep learning and multi-layer feature map fusion has the characteristics of high accuracy and fast retrieval.(2)Image Retrieval Based on Convolution Feature Fusion and Sensitivity DetectionBased on a deep learning of image retrieval method,the features extracted usually contained the complicated background noises,which resulted in a low level of accuracy in image retrieval.The research focus of this paper is the emphasis of the methods used to extract the salient features of images.In the view of this problem,the methods of feature fusion and saliency detection are proposed in this paper.The method firstly trained deep convolutional neural network model used in image classification,and then fused the features of maps after image convolution in order to obtain the salient features of retrieved images.Finally,the retrieved images are calculated using the cosine distance of the salient features.The experiment shows that the methods used in this paper are able to effectively improve the accuracy of retrieval and that the robustness is relatively high,compared to the current mainstream methods.