An Investigation Of Fabric Texture And Color Analysis Based On Dual-Side Imaging Technology

Fabric texture and color analysis as a significant fraction of the textile manufacturing process, traditionally based on human eyes with the assistant of magnifying glass, which has the disadvantage of time consuming and low efficiency. In recent years, many researchers have attempted to develop some new digital and objective methods to analyze the fabric structure parameters based on single side image. However, the texture and color feature of the fabric with dual-side surfaces is impossible to be recorded completely through one-side image. Therefore, this paper proposed a method of analyzing the fabric texture and color features based on dual-side imaging technique.The paper includes some main contents: the development of dual-side image acquisition system for yarn dyed woven fabric; the automatic yarn density measurement of yarn dyed woven fabric; the texture structure classification of yarn dyed woven fabric; the color clustering analysis of yarn dyed woven fabric; the color clustering analysis of printed fabric.Firstly, a review of the automatic methods used for the fabric texture and color analysis developed in nearly 30 years is provided. These researches have been classified into several categories based on different methods. Meanwhile, both the merits and demerits of these methods have been summarized and discussed. Then, the establishment of the dual-side imaging system, the general working flow of the dual-side imaging and the illumination influence has been discussed in detail. The binarization processing, Sobel edge detection technique and Radon transform are used to extract the equilateral angular reference markers. And affine transform is used to make the dual-side images corresponding to each other accurately on the pixel level. Next, the disadvantages of the single side spectrum based method for yarn density measurement have been discussed and the dual-side fused spectrum based method for the yarn dyed woven fabric density measurement is introduced. Three image fusion methods are used to fuse the dual-side images. With the aid of Fast Fourier transform technology, the frequency points corresponding to the yarn periodical components are extracted. Using the inverse Fast Fourier transform, the reconstructed images of the warp and weft yarns are obtained. It is found that both the maximum and minimum fusion methods are obviously accurate than the average method, whic h could be used for the yarn density measurement of yarn dyed woven fabric. After that, the dual-side gray level co-occurrence matrix has been proposed to describe the characterization of the yarn dyed woven fabric image. Four relevant features, such as co ntrast, correlation, entropy and homogeneity have been extracted, which could be regarded as the input vectors of the BP neural network. Once the establishment of the BP neural network system has been completed, the texture structure could be analyzed and classified into different categories. Besides, the yarn dyed woven fabric image has been decomposed into three sub-images in different channels respectively. The median filters with different template sizes have been selected to process the sub- images. The filtered sub- images in RGB color space, reconstructed from the three sub- images, could be converted into Lab color format. Ultimately, the result of color segmentation and classification has been obtained based on the Lab color space using the K-means clustering algorithm. The kind and number of the dyed yarns could be determined. Furthermore, the color image of printed fabric has been digitalized through an imaging system with a self-developed sample holder based on one set of flat scanner. The image of the printed fabric could be decomposed into three sub- images in red, green and blue channels. Different template sizes of median filters have been applied to eliminate the noises on the three sub- images respectively. Subsequently, the printed fabric image could be converted into Lab color space. After that, a self-adaptive K-means algorithm has been used to carry out the color clustering of the printed fabric image for the purpose of color classification. The different colored texture patterns could be determined.