Research On Embroidery Image Retrieval Based On Convolutional Neural Network

With the development of modern Internet technology and the emergence of commercial applications,the image resources contained in the network are numerous.Traditional retrieval techniques require manual annotation,which means that it will take a lot of time,manpower and material resources;and single features such as color,texture,or shape cannot fully express the rich content contained in the image,especially for complex patterns Embroidery images,using only one attribute as a search condition,cannot achieve the desired effect.Qinghai has a rich intangible cultural heritage,and it is one of the traditional ethnic embroidery with distinctive ethnic characteristics,regional characteristics and rich heritage.The ethnic embroidery in Qinghai is rich in subject matter,colorful in color,wide in use,and of great artistic value.It is particularly important to keep its records in the form of digital images.Image recording means that a large number of embroidery image resources will be generated,which will involve resource search and other operations.Therefore,the use of these embroidery image resources to achieve targeted,efficient and accurate retrieval is of great significance for the convenience of related personnel to query and use,the efficient management of image resources and the protection of the inheritance of embroidery cultural heritage and the spread of national culture.Since the convolutional neural network has a strong processing capacity for images,it can not only directly take images as input,but also does not require additional image preprocessing and feature extraction,so this article aims at the application of convolutional neural networks in embroidery image retrieval Conducted in-depth research.For the retrieval of Qinghai national embroidery images,this article does the following related work:1.The VGG16 convolutional neural network model is adopted,and the embroidery image features are extracted and saved through a series of operations such as convolution,and the image feature vector space model is constructed;2.In order to effectively improve the efficiency of embroidery image retrieval and reduce the redundancy of image features,PCA principal component analysis technology is used to achieve the conversion of high-dimensional feature vectors to low-dimensional images to achieve dimensionality reduction,and build on this basis The feature vector mapping space;3.When performing embroidery image retrieval,the cosine similarity algorithm is used to compare and match the feature vector of the embroidery image to be retrieved with the feature vector of the feature library,and then the embroidery image retrieval;4.Finally,based on a series of work,an embroidery image retrieval system was designed.