Research On Clothing Image Retrieval Method Based On Hand-drawn Sketches

In recent years,with the wide application of e-commerce and touch-screen terminals,tens of thousands of products are searched on Taobao every day,among which clothing retrieval has become a focus of attention.The early text-based search method can no longer meet the needs of the public,and then,"search by image" search method has been rapidly promoted.At present,clothing retrieval is mainly carried out by taking photos on major e-commerce platforms,but not all users can take photos anytime and anywhere.In addition,photographs are greatly affected by lighting,scene and equipment performance,resulting in low retrieval accuracy and low efficiency.Hand-drawn sketch is one of the earliest forms of expressing ideas and communicating with each other,and its greatest characteristic is that it can express the shape of the object directly.Therefore,in order to solve the existing clothing image retrieval based on text and content with low accuracy and efficiency,this paper proposes a clothing image retrieval method based on garment salient area detection and hand-sketched sketch.At first,as current hand-drawn sketch can not be matched with the clothe image and the problem of low accuracy,a method of garment salient area detection and edge extraction based on regularized random walk is proposed in this paper.This method uses regularized random walk algorithm to get garment salient area,and then uses edge extraction algorithm to get garment edge information,and weighs the edge information according to garment salient area to get garment edge image.It can eliminate the interference of background information,save the outline feature information of clothing effectively,solve the difference between sketch and clothing image,and improve the retrieval accuracy.Then,in order to solve the problem of apparel variety and low matching efficiency,this paper proposes a clothing image retrieval method based on hand-drawn sketch and reordering.The distance correlation coefficient is used to measure the similarity distribution between the two images,and the images in the clothing image database are divided into two disjoint partitions.And then it divides the images in the retrieval list into three clusters according to the partition created in the previous step.Finally,update rules are defined to update the similarity scores as the final retrieval results.The experimental results show that the proposed method improves the accuracy and efficiency of clothing retrieval,and is suitable for clothing retrieval based on hand-drawn sketch.Finally,based on the above methods,combining with the current actual application scenarios and user needs,the clothing retrieval system framework based on hand-sketched sketch is proposed,and the prototype design of the system is implemented.The system is easy to operate,and of high reliability and good robustness.