Research And Implementation Of Sketch Retrieval Method Based On Deep Learning

With the rapid growth of the amount of image data on the Internet,the difficulty of image retrieval is further increased,which also poses a greater challenge to the effectiveness of image retrieval methods.Compared with the mainstream text-based or content-based image retrieval methods,due to the better intuitiveness of hand-drawn sketches,the sketch-based image retrieval method also has strong practical application value in image retrieval,especially in fine-grained image retrieval.Hand-drawn sketches are only composed of contour lines,lacking details such as color and texture,and showing a large difference in feature distribution from images,so sketch-based image retrieval is a typical cross-domain retrieval problem.This paper takes the sketch-based image retrieval problem as an entry point,and mainly studies the problem of extracting domain invariant features in common subspaces of different domains.The second is the retrieval efficiency under large-scale image data sets.The rapid growth of data volume makes image retrieval face problems such as storage space and calculation speed.In order to solve the cross-domain problem in sketch-based image retrieval,an adversarial network model is designed in this paper.The stepwise training of classification subnetworks and domain adversarial subnetworks ensures the inter-domain correlation and intra-domain discrimination of output features.In addition,the effects of different branch network models on retrieval accuracy are compared through experiments.In addition,in order to solve the problem of large-scale sketch-based image retrieval,a deep hash model was designed on the basis of the adversarial network model.By combining correlation learning and hash learning stages,an end-to-end hash network is realized,which improves the calculation efficiency of the model.Finally,the retrieval effect of the deep hash model is verified through multiple experiments.Finally,based on the above algorithm,a sketch-based image retrieval system was developed to visually show the improvement of the retrieval efficiency and accuracy of the algorithm.At the same time,the system is used to make a more comprehensive collection of information such as sketch data and drawing stroke order to facilitate further research.