Using Savitsky-Golay filter and interval curve fitting in order to hyperspectral data compression

Abstract

Hyperspectral (HS) Images because of simultaneous acquisition of data on hundreds of narrow and close spectral bands have a high degree of inter-band correlation. For this reason and the large volume of this type of data we have to compress them. There are various lossy / lossless compressing methods acting on spectral and / or spatial domain. In this paper a lossy compressing method is proposed based on rational function curve fitting of spectral reflectance curve (SRC) intervals of image pixels after smoothing themutilizing Savitsky-Golay (SG) filter. The results demonstrate good performance in comparison to the other methods of compressing, including principal components analysis (PCA) as a bench mark method in this area. The degree of S-G smoothing filter and its window length changes in order to achieve the best performance. Also, we propose a special method for dividing SRCs into a number of adjacent non-overlapping intervals. Each of these sections is fitted by a variety of rational fraction curves, individually. The best fitted curve coefficients will be used as the new data cube.