Fabric image production system taking account into human feeling. Texture Image Prediction from Kansei Properties with Neural Networks.

Abstract

In this paper, we discussed a method to quantify the relation between KANSEI properties and physical properties to design an artifact. As a case study, the relation between the visual impressions of fabrics for office chair and those physical properties were i nvestigated using a neural network model. The purpose of this paper is to predict the fabric structural parameters and produce a fabric image from visual impressions. First, to acquire input and output data for a neural network model, a subjective evaluation of fabrics and measurement of their physical parameters were carried out. Thirty-four fabrics were evaluated by 57 subjects using 20 rating scales with the semantic differential method. Four kinds of fabric structural parameters: weave structure, yarn thickness, yarn density ratio of warp to weft, and color of fabric surface, were measured. The relations between the visual impressions and the parameters were formulated using a multilayer neural network. The performance of the neural network model was evaluated by a trend evaluation and a subjective evaluation of predicted fabric surface. Obtained results are summarized as follows.(1) The geometric parameters of fabrics were strongly related to the following visual impressions: luxuriousness, grace, softness, wetness, modern sense, sharpness, roughness and warmth.(2) The color of the fabric is affected by its grace, softness, clearness, wetness, warmth, and sharpness.(3) The fabric images reflected to activity, modern sense, roughness and wetness of fabric can be produced using the neural network. Finally, the system to produce the fabric image from visual impressions was proposed as an application of the neural network. Although th e system predicts a fabric image in the limited region, the system suggests a possibility of design of an artifact that reflects KANSEI properties quantitatively.