2017 Fiscal Year Annual Research Report
Single-shot Hyperspectral Fluorescent Imaging
| Project/Area Number |
16K16095
|
|---|
| Research Institution | National Institute of Informatics |
|---|
Principal Investigator |
鄭 銀強 国立情報学研究所, コンテンツ科学研究系, 助教 (30756896)
|
|---|
| Project Period (FY) |
2016-04-01 – 2018-03-31
|
| Keywords | Fluorescent Imaging / Reflective Imaging / Multispectral Images / Image Separation / Hyperspectral Recovery |
|---|
| Outline of Annual Research Achievements |
Following the researches of FY2016 that focus mainly on fluorescence/reflectance separation using hyperspectral images, the researches conducted in FY2017 are based on multispectral images, especially ordinary RGB images. The following two aspects have been investigated. (1). Using a set of RGB images captured under narrow band illuminations for fluorescence/reflectance separation. Since multispectral images, especially RGB images, are much easier to capture than hyperspectral images, we have considered to use multispectral imaging for fluorescence and reflectance separation. A robust separation algorithm was developed, together with its accelerated variant with known fluorescent emission spectra. The algorithms and experiment results interested a giant domestic maker, and we are collaborating on improving one fluorescent imaging product from this company using similar techniques. (2). We have also investigated the feasibility of reconstructing hyperspectral images from RGB images. A database including fluorescent materials was prepared, and a nonlinear mapping based image reconstruction method was developed. Compared with reflective scenes, we have found that the dimensionality of fluorescent scenes is higher, yet hyperspectral reconstruction from RGB images under a specific illumination type is still feasible.
|
