Research And System Implementation Of Thangka Image Retrieval Algorithm Based On Deep Learning

As an important part of Xixia art research,Heishui city thangka is an important part of Xixia art research.Due to the continuous increase in the number of unearthed,accurate classification and efficient retrieval of it have become an urgent task Intelligent retrieval technology based on the combination of deep learning and traditional retrieval algorithms provides powerful technical support for Thangka’s efficient and accurate retrieval and digital protection.The research object of this article is Heishuicheng thangka.The images in the thangka image database can be divided into four categories:Buddha,Bodhisattva,patron,and patriarch.As a cultural carrier rich in characteristic information,thangkas have complex texture information and richer semantic content.However,the number of thangka cultural relics that have been excavated is extrernely limited,so they are precious.How to complete the thangka image retrieval task with complex texture features under the support of extremely limited data samples is a difficult problem faced by this article.In response to this problem,this paper proposes traditional artificial modeling algorithms and two deep learning-based methods,aiming to fully extract and use the features of thangka images to complete the retrieval task of thangka images under limited feature samples.The main work of this paper is as follows:First,to solve the problem that there is no public Thangka image data set,this paper collects and preprocesses Thangka images through various channels,and creates Thangka image data set.Based on the created Thangka image data set,the SIFT algorithm is used for image retrieval,and the SIFT algorithm is improved,that is,the feature points extracted by SIFT are filtered by random sample consensus algorithm,and then the feature points are matched.Second,this paper further proposes two image retrieval algorithms based on convolutional neural networks,which are based on VGG16 and ResNet50.After that,PCA dimensionality reduction is performed on the features extracted by the two algorithms,and the retrieval experiment of thangka images is performed in different dimensions.The experimental data obtained is analyzed,and the algorithm based on ResNet50 has a better retrieval effect.And then,aiming at the problem of long time and low retrieval efficiency when using ResNet50 to retrieve,this article combines the image retrieval algorithm of hash algorithm to retrieve Thangka.The principle of this algorithm is to perform hash mapping on the thangka image features extracted by ResNet50,and then use Hamming distance to measure the similarity.Through the analysis of experimental data,it is obtained that the image retrieval algorithm based on the hash algorithm can improve the retrieval efficiency of thangka images,and it is the most suitable algorithm for this thangka retrieval.Finally,according to the thangka image retrieval algorithm proposed in this article,a thangka image retrieval system is designed and constructed,which can display and query the thangka image,and display the retrieval effect of the thangka image based on the algorithm.