2020 Volume 93 Issue 1 Pages 14-22
A method for evaluating the correlation between visual textures, i.e., convex-concave shape (surface texture), haze, gloss, brightness, of the specimen surface was proposed using self-organizing maps known as the neural network model of machine learning. It takes a lot of time and effort to weight the correlations in detail because there are many visual factors. Therefore, a simple comparison was verified using the similarity ratio of weighting pattern which appears in the output layer of self-organizing maps. First, the efficacy of this method was confirmed using the functions such as Y=X and Y=sin(X). Additionally, this method was applied to the visual texture evaluation of a transparent acrylic resin which had been shot blasted. As a result, it was shown that several surface texture parameters affected the haze, gloss and brightness of the specimen and that these parameters were maximum height Sz, maximum peak height Sp, root mean square slope Sdq and aspect ratio of texture Str.