Visual Features and Classification Based on Machine Learning for Yukatas, Aloha Shirts and Kariyushi Shirts

Abstract

  Image analysis was applied to evaluate the visual features of yukatas, aloha shirts, and kariyushi shirts with various colored patterns. The color measurements of the patterns were performed with a color scanner. L^*, a^*, b^*, C^*, and h of the CIE color system were obtained for each pixel (i,j) in the image. For color information, mean values of L^*, C^*, and h parameters for all pixels (AVE-L^*, AVE-a^*, AVE-b^*, AVE-C^*, and AVE-h) were obtained. For shape information, the angular second moment, contrast, correlation, and entropy parameters statistically extracted from gray-level distributions of color patterns were measured. The visual features were discussed in relation to similar and different properties of colored patterns. Furthermore, a forward-propagation neural network learning algorithm was applied and gave very good results in improving the accuracy of the visual evaluation system. The trained feedforward neural network model was successfully implemented to demonstrate the feasibility of the application of neural networks for the classification of yukatas, aloha shirts, and kariyushi shirts of varied colored patterns.