Navigation

Filter Curve Estimation of Multispectral Cameras

Filter Curve Estimation of Multispectral Cameras

Light spectra are an important source of information for many classification tasks. These range from applications in agriculture and medicine to tasks in forensics, e.g. to detect abnormalities that are not visible to the human eye:

Typically, hyperspectral cameras are used to capture light spectra. Unfortunately, they are very expensive and are not able to record videos. Thus, a cheaper alternative is highly desirable. This cheaper alternative could be a multispectral camera, which typically has around ten bands that wide ranges of the light spectrum by using wide-band filters. A typical imaging pipeline looks like this:

A reconstruction step is necessary to recover light spectra, since a single multispectral channel cannot be connected to a single wavelength. For this reconstruction, it is necessary to know the spectral behavior of the filters. Unfortunately, the filter curves provided by the manufacturers are often inaccurate, which severely degrades the reconstruction result.

The task of this master thesis is to estimate the filter curves based on a measured spectrum of a source of illumination and known reflected light spectra from a color chart. First, a simulation environment has to be set up. Afterwards, the developed algorithms shall be tested with a real-world multispectral camera.

Vorkenntnisse

  • Mathematics
  • Experience in Python

 

Betreuer

Frank Sippel, M.Sc.
frank.sippel@fau.de

Raum 06.0192
Cauerstr. 7
91058 Erlangen

Hochschullehrer

Prof. Dr.-Ing. André Kaup
andre.kaup@fau.de

Raum 06.031
Cauerstr. 7
91058 Erlangen