A non-parametric smoothing method is presented that reduces noise in multi-wavelength imaging data sets .
Using Principle Component Analysis ( hereafter PCA ) to associate pixels according to their ugriz -band colors , smoothing is done over pixels with a similar location in PCA space .
This method smoothes over pixels with similar color , which reduces the amount of mixing of different colors within the smoothing region .
The method is tested using a mock galaxy with signal-to-noise levels and color characteristics of SDSS data .
When comparing this method to smoothing methods using a fixed radial profile or an adaptive radial profile , the \chi ^ { 2 } -like statistic for the method presented here is smaller .
The method shows a small dependence on input parameters .
Running this method on SDSS data and fitting theoretical stellar population models to the smoothed data of the mock galaxy and SDSS data , shows that the method reduces scatter in the best-fit stellar population analysis parameters , when compared to cases where no smoothing is done .
For an area centered on the star forming region of the mock galaxy , the median and standard deviation of the PCA-smoothed data is 7 Myr ( \pm 3 Myr ) , as compared to 10 Myr ( \pm 1 Myr ) for a simple radial average , where the noise-free true value is 7.5 Myr ( \pm 3.7 Myr ) .