Subjective Evaluation of the Predictive Performance and Operability of a Two-Layer Regression-Based Texture Mixing...

This paper proposes and evaluates a texture mixing system for user-driven texture creation in virtual environments. The system is based on our previously proposed two-layer regression model. The first layer serves as a texture model that maps user actions to vibration features, whereas the second layer estimates the first-layer model parameters from texture features. In this study, we improved the dataset and training strategy to enhance extrapolative performance and implemented the resulting model as an interactive texture mixing system. Objective spectral similarity and subjective similarity evaluations revealed that the reproduction performance of the proposed model was comparable to that of a data-driven reference model for approximately 60% of the textures. We also showed that, under simplified single-peak conditions, nonexpert users could adjust texture-feature sliders to approach target textures within approximately one minute and with reasonably high accuracy.