Textile image retrieval using joint local PCA-based feature descriptor

Research output: Chapter in book / Conference proceedingChapter in an edited book (as author)Academic researchpeer-review


In apparel product development process, sales and merchandisers of apparel companies always spend a lot of manual effort and time to search the fabric swatch with exact or similar pattern required by their clients of buying houses in the physical fabric swatch hangers or fabric swatches database with targeted pattern and color. Meanwhile, content-based image retrieval is a research focus of image processing. How to extract feature conveyed the details information of the image is very important. However, the existing feature descriptor for the content-based image retrieval usually extracts the histogram feature of the image which cannot convey any detail information of the image. In order to overcome this problem, we propose a novel feature descriptor based on the classical unsupervised feature extraction algorithm: principle component analysis (PCA) which can automate the fabric swatch searching process for apparel companies. In the proposed method, the textile image is segmented into small block images. Based on the small block images, we extract 2D color features from R, G, and B channels, respectively, and then extract 2D orientation feature based on the orientation image. In contrast to histogram features, the extracted 2D color feature can keep the color detail information of the image, and the 2D orientation can obtain the edge information of the image. At last the retrieved images based on a novel joint feature classification criterion can be found. Experimental results clearly demonstrate that the feature extracted by the proposed PCA feature descriptor outperforms the histogram features. Furthermore, the user can highlight the retrieval content by selecting relevant 2D feature for textile searching.
Original languageEnglish
Title of host publicationApplications of Computer Vision in Fashion and Textiles
PublisherElsevier Inc.
Number of pages19
ISBN (Electronic)9780081012185
ISBN (Print)9780081012178
Publication statusPublished - 10 Oct 2017


  • Content-based image retrieval
  • Histogram feature
  • Sparse representation
  • Supervised learning
  • Unsupervised learning

ASJC Scopus subject areas

  • Engineering(all)
  • Materials Science(all)

Cite this