05月24日(Wed) 15:50〜17:30 L会場(ウインクあいち-10F 1003会議室)
| 演題番号 | 2L3-OS-09b-1 |
|---|---|
| 題目 | DCNNを用いた画像の質感認知ー音象徴性からのアプローチー |
| 著者 | 權 眞煥(電気通信大学情報理工学研究科情報学専攻) 川嶋 卓也(電気通信大学情報理工学部総合情報学科) 下田 和(電気通信大学 大学院総合情報学専攻) 坂本 真樹(電気通信大学大学院情報理工学研究科総合情報学専攻) |
| 時間 | 05月24日(Wed) 15:50〜16:10 |
| 概要 | Material (or texture) perception represents that various physical quantities belonging to objects are detected by human sensory receptors and processed in the brain. However, it is difficult to define material perception because the amount of information is enormous in material and texture. In recent years, some researchers have raised a possibility that perceptual characteristics of textures can be integrally expressed by sound symbolism. Sound symbolism represents a phenomenon in which a certain amount of information perceived from the physical world is strongly associated with phonological elements in the brain. In this research, we aim to generate material and texture expressions using sound symbolic words as variables to converge various material and texture features and Deep Convolutional Neural Network (DCNN). As a result, it became possible to output stochastic phoneme of sound symbolism to the input image. Our achievement is expected as a new method capable of expressing various fine texture information comprehensively. |
| 論文 | PDFファイル |