Recently, an iterative optimization method was proposed that determines the spectral transmittance of a color filter which, when placed in front of a camera, makes the camera more colorimetric [1]. However, the performance of this method depends strongly on the filter (guess) that
initializes the optimization. In this paper, we develop a simple extension to the optimization where we systematically sample the set of possible initial filters and for each initialization solve for the best refinement.
Experiments demonstrate that improving the initialization step
can result in the effective ‘camera+filter’ imaging system being much more colorimetric. Moreover, the filters we design are smoother than previously reported (which makes them easier to manufacture).