Design And Development Of Printed Fabric Online Retrieval System

Printed fabrics,as the raw materials of clothing and home textiles that can reflect the trend of the times,have developed particularly rapidly in recent years.The production proportion of printed fabrics in fabric enterprises is also increasing.The patterns of printed fabrics are complex and varied,and the types are countless.The diversification of fabric patterns has caused great pressure on the production and management of fabrics,such as high labor intensity in imitation production,long production cycle,high inventory management costs,etc.How to solve the above problems,promote the transformation and upgrading of enterprises,and enhance the competitiveness of enterprises is an urgent problem for traditional fabric manufacturers to cope with the development of the times.The traditional real sample storage method is inefficient,the storage cost is high,the single text-based image retrieval is subjective,and the labor cost is high.The single content-based image retrieval needs to provide the image to be inspected.Therefore,this paper studies the content based image retrieval technology of color similarity and texture similarity,and combines text-based image retrieval with contentbased image retrieval to meet the actual retrieval needs of enterprises,It is integrated into the system to realize the online retrieval of printed fabrics.The main work of this paper is as follows:(1)Based on the results of literature research and the requirements of textile enterprises for fabric retrieval system,the online retrieval system for printed fabrics is analyzed and designed according to functional requirements and performance requirements,including the establishment of feature index library and the design of retrieval module.At the same time,2000 original images of different printed fabrics are selected according to the characteristics of printed fabrics.After image preprocessing,the material library is expanded.After cutting,rotating,scaling and other operations,10 sub images can be generated from each original image to build a data set of evaluation and retrieval performance with 20000 fabric images.(2)A fabric image retrieval method focusing on pattern texture similarity of printed fabrics is proposed.The SURF feature of printed fabric is extracted to represent the pattern texture feature of printed fabric.In order to improve the problem of large feature dimension and complex calculation,VLAD coding and PCA dimension reduction are introduced to construct a texture feature with rotation and scale invariance to represent the pattern texture of printed fabric.In terms of similarity measurement,Ball tree is used to build an index.Through experimental verification,the average retrieval accuracy m AP reaches 83.5%,and the precision P@10 95.5%,recall rate R@10 59.2%,and the average retrieval time is 0.488 s.The effectiveness of the proposed method is further verified by comparing different texture feature extraction methods.(3)A fabric image retrieval method focusing on color similarity of printed fabrics is proposed.In order to solve the problem that common color representation methods are difficult to characterize due to the wide variety and distribution of printed fabric colors,a method combining two color representation algorithms was designed.A method of image dominant color based on color quantization is proposed.Weighted aggregation method is used to avoid the color with small missing part when representing dominant color.The regional color moments are proposed to represent the global color features of the image,which enhances the expression ability of color spatial distribution.The two methods are fused by multiplying similar shapes to form complementarities.Through experimental verification,its precision P@10 Up to 85.1%,recall rate R@10 It reaches 44%,the m AP reaches 76.2%,and the average retrieval time is 2.3s.The effectiveness of the proposed method is further verified by comparing different color feature extraction methods.(4)This paper proposes a retrieval strategy that combines text-based image retrieval technology with content-based image retrieval technology.In order to solve the problem of image retrieval when there is no fabric image to be inspected in actual production,this paper proposes to combine text-based image retrieval technology with content-based image retrieval technology.Text based image retrieval only needs to provide simple text retrieval to realize rough fabric retrieval,while content-based image retrieval realizes accurate fabric image retrieval.Text retrieval uses TF-IDF algorithm to extract keywords,Content based image retrieval uses the algorithm proposed in this paper,and the results show that this strategy has achieved good retrieval results.Finally,the method and retrieval strategy designed in this paper are embedded into the printed fabric online retrieval system developed by using the Django+Vue framework to separate the front and rear ends,and the algorithm is applied in practice.