Research And Implementation Of Personalized Sketch-Based Image Retrieval Using Deep Learning

With the rapid development of big data and artificial intelligence technology,more and more computer vision problems have been solved by means of deep neural networks.Sketch-Based Image Retrieval is an important branch of image retrieval.In recent years,studies have also shown that deep learning can improve the accuracy of retrieval.In the traditional sense,the sketch-based image retrieval mostly performs feature calculation and similarity matching according to the rules defined by human-beings.The retrieval results are generally similar in contour,and the complete semantic information of the image is missing.Simultaneously,due to the inherent ambiguity of the hand-drawn image,there is a semantic”one-to-many"category mapping relationship between the hand-drawn image and the natural image.From the user's point of view,the same hand-drawn image may represent the semantics of many different things.In addition,the user's drawing level has many different characteristics,so the search results generally cannot fully meet the needs of users.Based on the above challenges in the field of sketch-based image retrieval,this paper proposes a kind of deep full convolutional neural network structure suitable for the training of general model of sketch-based image retrieval.The model has strong generalization ability and testing on the Flickr15K dataset achieved higher mean average accuracy in recent years.On the basis of the pre-trained general model and the result images selected by the user feedback,we constructed the personalized model training dataset and combined the user history feedback image with the input hand-drawn image as input.Then we use the transfer learning idea to fine-tune the general model parameters so that the neural network can achieve fine-grained image semantic feature learning.This is the first time that we propose to solve the problem of personalization in the field of sketch-based image retrieval by the idea of transfer learning.Experiments show that the general model can realize fine-grained image semantic feature learning after migration,so as to meet the personalized retrieval requirements of user input hand-drawn images.Through the above research,the paper presents a complete solution and an innovative application algorithm model for the user's personalized sketch-based image retrieval.