Pattern‐driven color pattern recognition for printed fabric motif design

Abstract

Pattern-driven design method is an important data-driven design method for printed fabric motif design in textiles and clothing industry. We introduce a novel framework for automatic design of color patterns in real-world fabric motif images. The novelty of our work is to formulate the recognition of an underlying color pattern element as a spatial, multi-target tracking, classification, segmentation and similarity association process using a new and efficient color feature encoding method. The proposed design method is based on pattern-driven color pattern recognition and indexing from the element image database. A series of color pattern recognition algorithms are used for color and pattern feature extraction. The local statistical corner features and Markov random field model are used for motif unit tiling detection and conversion. The color feature encoding problem is modeled in a gray-scale color difference optimization problem, which can be solved quickly by existing algorithms. Color pattern feature matching, segmentation and indexing techniques are then used to locate and replace the elements in the motif unit image with similar elements in the database. Experiments show that the approach proposed in this study is effective for color pattern recognition and printed fabric motif design.

A new method for textile pattern recognition and recoloring

“…Image pattern recognition based on low‐level feature is usually performed on one dimensional color channel 8 . Various color‐encoding methods have been developed to obtain a 1D scalar representation of the 3D vector color space, and the methods can be divided into two categories according to the feature extraction characteristics: local and global conversion.…”

“…Image pattern recognition based on low-level feature is usually performed on one dimensional color channel. 8 Various color-encoding methods have been developed to obtain a 1D scalar representation of the 3D vector color space, and the methods can be divided into two categories according to the feature extraction characteristics: local and global conversion. Since local conversion methods have the shortcoming that conversion of a fabric image is not homogeneous and may produce halo artifacts, 10 this study uses a global conversion method based on Euclidean-Log metrics 11 for preserving structure and contrast in the fabric image, 10 which is expressed as:…”

“…It is helpful to remove the color mixture phenomenon on homogeneous yarn. Zheng 8 proposes a framework for automatic recognition and generation of color patterns from the given fabric motif images. The novelty of the framework is to formulate the recognition of an underlying color pattern element as a multitarget tracking, classification, segmentation, and similarity association process through an effective color feature encoding method.…”