Filter selection for multi-spectral image acquisition using the feature vector analysis methods

The selection of filters plays a fundamental role for the setup and the performance of a multispectral image acquisition system. When the number of available filters is large, the full search method becomes computationally expensive. In this paper the filter selection problem is defined as a dimensionality reduction process by selecting the filters that contains most of the essential information, using the same criteria as the Principal Component Analysis. The proposed method is also expanded by processing multiple filter estimations to improve the stability and the quality of the results. The filter selection methods were evaluated under the CIE standard illuminants A and D65 using 160 filters and a spectral reflectance database of 1269 curves.