Research On Image Retrieval Methods For Footwear Products

The field of e-commerce has also gained unprecedented vitality with the fast development of mobile Internet,and various large and medium-sized e-commerce websites have emerged one after another.However,there are many shops in the online shopping mall and there are many kinds of goods.The number of pictures used to show shoe products is also increasing in geometric multiples.There are large or even super-large capacity shoe image databases,but a large number of footwear images cannot be effectively accessed and used.How to quickly and effectively retrieve footwear image data from massive image resources and how to effectively utilize massive shoe images have become urgent problems to be solved.Therefore,in order to solve the problems that the existing shoe image retrieval seldom uses semantic attributes and the retrieval objects are limited to coarse grains,which leads to low retrieval accuracy,the research of shoe image retrieval method based on component detection and semantic network is proposed to realize and improve the accuracy and precision of shoe image retrieval,so as to better meet the needs of practical application.To resolve the problem of lack of key semantic attributes or inaccurate attribute division in the existing shoe image retrieval methods,a retrieval method by part aware and attribute learning is proposed for various shoe images.The method firstly carries out part detection on the shoe image to be detected and the labeled data set input in the shoe image database to obtain the detected part areas respectively;Secondly,four underlying features of HOG,LBP,SIFT and color histogram of the detected part region are extracted and matched with semantic attributes.Then,through attribute learning and training LSPC classifier,attribute value vectors of the image to be detected and the data set are obtained respectively.Finally,the similarity between the attribute value vector of each candidate image in the data set and the attribute value vector of the image to be detected is calculated,and the retrieval results are output in the order of high and low.To resolve the problem of traditional methods typically retrieve similar replicated images,which are primarily based on large-scale coarse-grained retrieval,but with lower precision.Fine-grained image retrieval is positioned in the fine-grained image,identification and retrieval subclasses,and it is widely used in image retrieval and image analysis of fine-grained.The traditional image retrieval task only extracts the coarse-grained features of the image,but it cannot be effectively used for fine-grained retrieval.And it lacks key semantic attributes that make it difficult to distinguish the nuances between the parts,a fine-grained image retrieval method by part detection and semantic network is proposed for various shoe images.Firstly,the part-based detection was conducted to detect the undetected shoe images combing the annotated training dataset of shoe image.Secondly,the semantic network is trained based on the semantic attributes of the detected shoe image and the training image,and the feature vector are extracted respectively.Then,the principal component analysis is used for dimensionality reduction.Finally,the results were implemented and output by the metric learning to calculate the similarity between images,and fine-grained image retrieval is implemented.Finally,according to the actual application scenarios and user requirements,the above algorithms are combined to design and implement a shoe image retrieval system.Through the analysis of system requirements,the prototype system realizes three main basic functions: feature extraction of shoe images,retrieval of shoe images and display of retrieval results.The main interface of the system consists of three parts: user operation part,image display part to be inspected and result image display part.