Visual Features of Colored Patterns of Cotton, Silk and Wool Fabrics and Distinction of Fiber Materials Based on Mechanical Learning

Abstract

Color feature analysis was applied to evaluate visual features of cotton, silk and wool fabrics with various colors and patterns by using CIELAB color space. Color measurements of these fabrics were done with a commercially available color scanner. L*, a*, b*, C* and h of CIE color system were obtained for each pixel (i, j) in the image. Mean values for L*, C* and h of all pixels (AVE-L*, AVE-a*, AVE-b*, AVE-C* and AVE-h) were obtained as color information parameters. The angular second moment (ASM), contrast (CON), correlation (COR) and entropy (ENT) statistically extracted from distributions of L* for each fabric, were measured as shape information parameters. The visual features were discussed in the relation to similar and different properties of colored patterns. Furthermore, the high performance neuron training algorithm using mechanical learning was introduced to tune the network to maximize accuracy of the description of the visual evaluation system. This feedforward neural network training algorithm gives much better results. The trained feedforward neural network model was successfully implemented to show the feasibility of neural network applications for distinction of fiber materials using colored patterns of fabrics.