Design And Development Of Recognition And Retrieval System For Artworks Based On Image Feature

Cultural relics and artworks are the crystallization of human civilization.Thousands of years of human history have left valuable art resources,and the classification and retrieval of art objects is an important subject in the field.Not only ordinary people have difficulty in mastering their content,but it is also difficult for professionals to master whole details of their information.It is an important trend to recognize and realize fast retrieval based on the image content of artwork.At present,artificial intelligence(AI)develops rapidly,image recognition and retrieval based on intelligent technology is the research hotspot.In this paper,AI technology is introduced into the classification and retrieval of artworks.Research contents and achievements are as bellow:(1)A system solution is put forward based on image content for recognition and retrieval of artworks,which is born from demand analysis,conceptual design of the database,detail discussion of architecture and functions of system.(2)The design details of the crawler program,which grabs pictures and texts of cultural relics and artworks from internet,are given in this paper.On the basis of the crawler,a data set is completed which has about 10 thousand images.Besides,the class imbalance state of the labeled images is analyzed,which can contribute to the selection of tricks for class imbalance.(3)The study details of the recognition and retrieval methods based on image content of artworks are given in this paper.Experiments on several data sets are implemented to test the performance of visual word and CNN model while describing image content and a further comparison is made among different convolutional models.Besides,oversampling is tested in class imbalance situations.Experimental values of AP,AUC,TopK accuracy and mAP show that VGG-16 is the most appropriate model to describe the visual content in recognition and retrieval task of the project in this paper.(4)A further exploration on the optimization of image recognition and retrieval algorithms is given in this paper.Two innovative algorithms are proposed for recognition and retrieval respectively,one of which is to promote the performance of recognition in class imbalance situation with category centers of CNN features,the other is to enhance the query image's semantic content to improve results in image retrieval tasks.Both algorithms are tested on Cifar-10,Cifar-100 and the dataset of artwork images.Experimental results show that both algorithms have obvious optimization performance and generalization ability.The former has significantly improved the AP value of image classification in class imbalance situation,and the latter has significantly improved both TopK precision and mAP in retrieval tasks.(5)The development details of the software platform of the proposed system of recognition and retrieval for artworks are given in this paper,to which the methods and algorithms explored in this paper are applied.The design and development of the server,the web client and the corresponding tests are all included in the development details.The technique of image feature analysis is introduced and applied to the classification and retrieval of artworks in this paper.The developed system has the function of recognition and retrieval based on image of artwork and the function of importing and uploading pictures and texts massively,which is extremely valuable and significant to acquire information of cultural relics and artworks.