Clothing Image Retrieval Based On Deep Learning And Perceptual Hashing

In recent years,with the rapid development of e-commerce,more and more users purchase clothing through the Internet.How to retrieve suitable clothing from a large number of clothing has become a hot research direction.However,the traditional text-based information retrieval of clothing products has disadvantages such as unclear description and unobjective text description,so the classification and retrieval of content-based clothing images has become a very significant subject.The clothing image is mainly divided into three major characteristics of style,color and texture,and the style,which is one of the main characteristics of the clothing image,is the focus of this paper.Clothing images have problems such as complex backgrounds,multi-scales,and multiple labels.Therefore,traditional methods for extracting clothing style features have certain limitations.This article will use the target detection network to extract the characteristics of clothing images,and then identify the clothing attributes.Retrieve similar clothing images and feed them back to the user.The following tasks have been completed:First,construct a clothing-oriented image data set,and transfer the trained model of the existing clothing data set to a self-built clothing data set.There are many existing clothing datasets,but the features provided by these datasets are not suitable for clothing image retrieval based on style features.Therefore,this article has built a clothing dataset with style features.This dataset consists of clothing sample dataset and clothing.The retrieval data set consists of two parts.In the paper,the COCO dataset is used to train the network model,and then the model is transferred to the self-built clothing sample dataset,thereby improving the training efficiency.Second,the target detection network and BP neural network are used to extract the multigranularity features of the clothing image.In this paper,the Mask R-CNN network is used to segment the clothing images to obtain clothing contours.At the same time,coarse segmented feature extraction is performed on the segmented clothing contours to obtain the basic categories of clothing images.Due to the highly deformable appearance of clothing,in order to extract more clothing details and improve the accuracy of clothing style recognition,the paper proposes a genetic algorithm to optimize the BP neural network,and uses the optimized BP network to extract clothing contours.Fine-grained features.Finally,the image similarity calculation strategy was used to retrieve clothing with similar attributes in the clothing retrieval database.According to the style characteristics of the extracted clothing image,the paper proposes a perceptual hashing algorithm based on the combination of singular value decomposition(SVD)and discrete wavelet transform(DWT)to calculate the similarity of clothing images,and returns a high similarity based on the calculated results.Clothing image.The analysis of experimental results shows that the multi-granularity feature extraction method based on clothing styles proposed in the paper has higher accuracy in identifying clothing styles,and its performance is better than the existing clothing image feature extraction methods;compared with existing clothing image retrieval algorithms Using the proposed perceptual hashing algorithm for clothing image retrieval has high accuracy and strong robustness.