Research On Image Retrieval Based On Multiple Features Fusion

With the emergence of Internet image information,image retrieval technology has become one of the research focuses on the field of computer vision.In image retrieval,different visual features can express different image contents.Low-level visual features can represent information like human visual perception,but they can’t express high-level semantic information,thus limiting the scope of use of this feature.In recent years,deep learning technology has developed rapidly,especially convolutional neural networks,which can extract deep features and can well express high-level semantic information.However,whether it is low-level features or deep features,the representation ability of a single feature is not comprehensive.By combining multiple features and describing the image content from different perspectives through complementary features,this problem can be solved well.Therefore,this paper proposes an image retrieval method that fuses multi-level features and an image retrieval method that fuses multi-model features.They can combine the advantages of multiple features and describe image content from different perspectives to improve the performance of image retrieval.The main contributions of this paper are as follows:1.To solve the problem that a single feature doesn’t adequately describe the image content,this paper proposes an image retrieval method that fuses deep features and Tamura texture information.Firstly,the Tamura texture information is extracted from deep feature,and deep feature is fused with the texture information to obtain the deep Tamura feature.This feature not only has high-level semantic information,but also contains low-level texture attributes.Secondly,this paper proposes a spatial layout optimization method with deep Tamura features,which can remove the interference of background noise and enhance the feature response intensity of the target object.Finally,the optimized deep Tamura features are normalized and whitened to obtain a robust image representation.Our proposed multifeature method can fuse different levels of visual features,which fully leverage the advantages of deep features and texture information,thus improving the accuracy of image retrieval.2.To combine the feature description ability of different network models,this paper proposes an image retrieval method based on feature fusion of multiple models.Firstly,different convolutional neural network models(VGG16 and VGG19)are used to extract the deep features,and then feature selection is carried out on deep feature to remove irrelevant feature maps.Secondly,spatial optimization and channel aggregation are carried out to improve the feature response of each feature map,thus highlighting the information of the target object.Finally,the enhanced deep features are integrated,which this method can combine deep features of different models to fully leverage the advantages of each feature and improve the performance of image retrieval.We conduct experiments on five public benchmark datasets.The experimental results show that the proposed method has better retrieval performance than some existing advanced image retrieval methods.Our method not only take full the advantages of different levels of visual features and deep features of different models,but also can enhance the ability of feature description and improve the accuracy of image retrieval.