Research On Image Retrieval Technology Of Ethnic Costumes Based On Semantic Segmentation

Costume culture is an important manifestation of the cultural connotation of various nationalities.With the development of network technology and multimedia technology,the accurate retrieval and application of digital image information resources is undoubtedly an effective means for the protection and inheritance of ethnic costume culture.However,due to the diverse colors,complex texture styles,and rich semantic features of ethnic clothing images,the current "search for images with pictures" direct retrieval method for ethnic clothing retrieval tasks often ignores the clues of local semantic features of clothing.The effect is not ideal.Aiming at the abovementioned problems,this paper studies the retrieval methods of ethnic costumes from the perspective of semantic feature fusion and hierarchical retrieval.The main work is as follows:(1)The construction of national costume image resource library.This article focuses on the two ethnic groups of Wa and Hani who live in Yunnan Province as the research objects.Through field data collection,sorting and preprocessing,a national costume image resource database containing multi-attribute information such as national costume categories and semantic tags of ethnic costume parts is established.(2)Semantic segmentation of ethnic costume images.In order to improve the speed of existing semantic segmentation methods while maintaining segmentation accuracy,this paper constructs a semantic segmentation model of ethnic clothing based on Deep Lab V3+.Through experiments and comparative analysis of the segmentation effect under different network structures,the results show that the improved semantic segmentation model can maintain a high segmentation accuracy,and also has a good performance in improving the running time.(3)Hierarchical retrieval of ethnic costume images based on the fusion of semantic features.First,this paper designs and trains a deep convolutional neural network to extract clothing images and semantically segmented clothing component features,and then integrates the overall global features of clothing with the local features of clothing components,and finally adopts a hierarchical retrieval strategy,first classification and then retrieval Complete the task of retrieval of national costume images.The experimental results show that the proposed retrieval strategy can effectively improve the retrieval accuracy of ethnic clothing images.(4)The design and implementation of ethnic clothing image retrieval system based on semantic segmentation.This paper uses Python+Tensor Flow+Tkinter to develop an image retrieval system for ethnic clothing.The system integrates functions such as semantic segmentation,feature extraction,feature fusion,and image retrieval,provides visual display of clothing semantic segmentation results,and has good interactivity.