A real-time spectral decomposition of streaming three-band image data is obtained by applying linear transformations. The Principal Components (PC), the Maximum Autocorrelation Factors (MAF), and the Maximum Noise Fraction (MNF) transforms are applied. In the presented case study the PC transform appears to provide the best result for separating signal from noise. The main difficulty for the more advanced methods is to obtain good estimates for the correlation structure of the noise since problems arise due to spatial and temporal coding of the video signal. A new MNF transform is proposed that utilised information drawn from the temporal dimension instead of the traditional spatial approach. Using the CIF format (352x288) frame rates up to 30 Hz are obtained and in VGA mode (640x480) up to 15 Hz.
|Title of host publication||Proc. 10th Danish Conference on Pattern Recognition and Image Analysis|
|Publisher||Department of Computer Science, University of Copenhagen (DIKU)|
|Publication status||Published - 2001|