Semantic Completion Annotation Of Ancient Architecture Based On Attention Mechanism And Formal Concept Analysis

As an important carrier of traditional culture,Chinese ancient architecture inherits long historical civilization.With the rapid development of technology and economy,more and more users beginning to upload and share images of ancient buildings on the Internet.Facing the increasing image resources of ancient architecture on the Internet,how to quickly select favorite images through accurate and complete semantic keywords has become an urgent problem to be solved in image annotation of ancient architecture.Effective feature extraction is the key to improving the annotation accuracy of ancient architectural images.However,although the current label complete annotation method based on convolution neural network can effectively establish the mapping relationship between image features and semantics,due to the obvious difference of contour boundary features in ancient architecture images and the spatial correlation between architectural components,only using convolution neural network can not mine the correlation between salient features and corresponding semantic of the image.Meanwhile,there is an implicit correlation between semantic labels of ancient architecture,and different tags play different roles in reflecting the characteristics of ancient architectural styles.Visual attention mechanism can pay more attention to the important features of the image and suppress the unimportant features.It can also allocate resources to better obtain image visual information.Formal concept analysis is an effective semantic analysis method.Therefore,to effectively enrich the semantics of ancient architecture images,this paper studies the complete annotation of the images based on visual attention mechanism and formal concept analysis.The main research work is as follows:(1)Aiming at the characteristics that different ancient architectural images have different outline features and the architectural semantics are interrelated,an image of ancient architecture complete annotation algorithm based on visual attention mechanism and semantic analysis of concept lattice is proposed.Firstly,the visual attention mechanism is combined with convolutional neural network to extract salient features such as the contour boundary of the image to be marked,and then an initial label set of the image is obtained by softmax classifier.Calculate the visual similarity of images to obtain the nearest neighbor image set of the image to be annotated.Secondly,use the concept lattice which is constructed by image nearest neighbor set and tag set as the input of formal background to analyze the semantic labels of ancient architecture images hierarchically and the candidate label set is output.Finally,the experiment results on the ancient architecture dataset shows that this method can effectively improve the recall rate of ancient architecture image tags and enrich the semantics of the images.(2)Different semantic labels of ancient architecture play different roles in reflecting the style characteristics of ancient architecture,furthermore,it may not effectively reflect the style characteristics of ancient architecture only according to the frequency of labels in the process of ancient architecture image annotation.To acquire the representative semantic labels of the image,a semantic complete annotation method of ancient architecture image based on attention mechanism and weighted concept lattice is proposed.Firstly,the initial label set is obtained by visual attention mechanism and convolution neural network;Then,the nearest neighbor image set and nearest neighbor label set are obtained through visual similarity,and the weight value of each label in the nearest neighbor label set of ancient buildings is calculated by information entropy to identify its role in the style characteristics of ancient buildings;Secondly,the weighted concept lattice is constructed based on the neighbor image and the neighbor label with weight.Finally,the weighted concept lattice is used to analyze the semantics of ancient architecture to obtain the label candidate set,and achieve complete labeling of images.The experimental results on the ancient architecture dataset demonstrate that the algorithm can improve the accuracy of ancient architecture image annotation to a certain extent.(3)A complete annotation system of ancient architecture image label is realized.A semantic complete annotation system of ancient architecture images based on visual attention mechanism and formal concept analysis is designed and implemented on the Matlab platform.