Development Of Image Retrieval System For Colored Spun Fabrics Based On Low-level Features

With their unique texture appearance,rich color changes,environmentally friendly production technology,and excellent product performance,colored spun fabrics are increasingly widely used in various premium apparel and home textile products.Due to the diversified development trend of colored spun fabrics,more and more fabric samples are stored in the sample libraries of many textile companies.How to find the required fabric data quickly has become an urgent issue for textile fabric manufacturers.At present,the common retrieval methods of colored spun fabrics by textile enterprises are manual retrieval or indexing by text and entering keywords for retrieval.These two methods are easily affected by subjective factors and are time-consuming and labor-intensive.With the increase in market demand,the abovementioned fabric retrieval methods can no longer meet the needs of enterprises,and the establishment of an image retrieval system for colored spun fabrics is an effective means to solve this issue.This paper applies content-based image retrieval technology to the image retrieval of colored spun fabrics,which can avoid the influence of human factors,cut down the production cycle of the company's products,achieve rapid response to all links in the production chain,improve production efficiency,and promote the digital and intelligent transformation of the firm process.The main research content of this paper is to establish a colored spun fabrics image retrieval system based on the characteristics of colored spun fabrics:(1)Convert colored spun fabrics into digital images that can be recognized and processed by the computer,and establish an image data set of colored spun fabrics.The Digi Eye image color management system was used to collect the image of colored spun fabrics.And then the images were cut,enhanced,and other pre-processing operations.Then the colored spun fabric image database is constructed according to the annotation and classification of the colored spun fabrics by the factory experts.A total of more than 6,000 images were obtained,including about2,000 stylized fabrics and about 4,000 solid-color fabrics.Solid-color fabrics were divided into9 categories according to their colors,and each category contains about 500 pieces.The colored spun fabric image database provides experimental samples for subsequent classification and retrieval experiments.(2)According to the characteristics of the colored spun fabric images,the color features and texture ones were extracted.The color feature extraction of the colored spun fabric images used color histogram,color moment,image dominant color.And the texture feature extraction used GLCM,LBP,Haar wavelet transform,Fourier transform.Through the retrieval experiment,the appropriate features are selected for the image retrieval algorithm design of the colored spun fabric.The experiment presents that the color feature performs better in the image retrieval of the colored spun fabrics compared with the texture feature.And each feature shows a certain difference in the image retrieval of different types of colored spun fabrics.(3)According to the difference in the performance of each feature in the retrieval of different types of colored spun fabrics,different retrieval algorithms are proposed for different types of colored spun fabrics,and SVM is used to classify colored spun fabrics.Construct two classifiers according to the label category of colored spun fabrics.The first one is the classifier I of solid-color fabric and stylized fabric constructed by the concatenation of LBP features and Fourier transform features.And the second one is the classifier ? of various types of solid-color fabrics constructed by using color moment features.After training,the average classification accuracy of classifier I reached 90.13%,and the average classification accuracy of classifier II was 84.80%.(4)A step-by-step search strategy for colored spun fabrics based on SVM classification is proposed.Namely,classifying firstly and then retrieving.First,the classification results are output in the form of probability to ensure the retrieval accuracy.And a threshold is set to determine the effectiveness of the classification.Then based on the performance of each feature in the image retrieval of different types of colored spun fabrics,the algorithm CMWT that combines color moment and wavelet transform is proposed for the retrieval of stylized fabrics,and the algorithm DCFT that combines the dominant color and the Fourier transform is used for solid-color fabrics.In the retrieval process,the corresponding algorithm is selected for image retrieval according to the classification results.The results show that the CMWT algorithm has a better retrieval effect in stylized fabrics,with the m AP reaching 82.56%;the DCFT algorithm has a better retrieval effect in solid-color fabrics,with the m AP reaching89.47%;and for the comprehensive image database,the DCFT algorithm has better retrieval results,with the m AP reaching 82.06%.The step-by-step retrieval strategy for colored spun fabric images based on SVM classification proposed in this paper can achieve higher retrieval efficiency without reducing retrieval accuracy.Compared with the algorithms DCFT and CMWT,the retrieval speed of proposed method is increased by more than 37%;compared with common retrieval algorithms for fabrics,the retrieval accuracy of the algorithm proposed in this paper is higher than other algorithms,with the m AP reaching 83.73%,indicating that the algorithm proposed in this paper is more targeted for the retrieval of colored spun fabric images.