Optimal Sampling and Principal Component Selections for Spectral Image Browsing

摘要

Spectral imaging is becoming popular. Spectral accuracy in measurements is an important factor, especially now when fluorescent and light emitting diode (LED) based light sources are becoming common. Browsing image sets in a modern network is also becoming relevant, but the problem with spectral data is that the file sizes are so large. An efficient compression method suitable for browsing purposes consists of principal component analysis with spatial subsampling. In this study, the optimal combinations of a sampling interval and parameters of the developed compression method are found for different data sets under several light sources. It is shown that depending on the light source, 3-20 nm sampling intervals are required. In addition, with different light sources and data sets, between three and six principal components must be used. With a suitable spatial subsampling mask, high compression ratios can be achieved with good results. The spatial subsampling is a fast operation and can be done online before transmission, which gives the client user a possibility to choose the compression ratio.