| （英） |
This work proposes a 3-Way Locally Structured Unitary Network (3-Way LSUN) as a volumetric data analysis method. With advances in measurement technology, the need for high-dimensional data analysis is increasing. Among high-dimensional data, there is a type of volumetric data that has information in three directions (vertical, horizontal, and depth) and can express three-dimensional objects, including contents. Because the amount of volumetric data is huge, processing and storage costs are high. Therefore, some dimensionality reduction is needed for efficient processing. Highly accurate analysis and reconstruction methods are required to reduce the dimensionality of the volumetric data. Block Principal Component Analysis (BPCA) is a typical example of an existing analysis method. However, dimensionality reduction using BPCA has issues with continuity between blocks, such as block distortion. Thus, this study extends the 2-Way LSUN proposed by Yasas textit{et al.} to allow for a three-direction analysis, and proposes a 3-Way LSUN as a highly accurate volumetric data analysis and synthesis transform. Through experiments on the dimensionality reduction and evaluation of the approximation error with the original, the significance of the proposed 3-Way LSUN is confirmed. The 3-Way LSUN is shown to achieve higher accuracy than the BPCA. |