Spectral Image Decolorization

We propose a novel method for decolorizing a color image to grayscale, which preserves the chromatic contrast to be lost if using a simple luminance mapping. Unlike previous work, the conversion performs in spectral domain after applying Fourier transforms on luminance and chromatic channels, and the resultant intensity is recovered from an inverse transform. Frequency spectra inherently provide the magnitude of changes (i.e. contrast) among the channels, in all spatial scales. Consequently, only simple arithmetic operations are needed to add chromatic differences to the final grayscale intensity, leading to realtime computational speed. Moreover, users can flexibly control the decolorization effects by interactively adjusting visually-intuitive parameters, in addition to our predefined values. The method has great potential to be incorporated in commercial software and printer drivers due to its simplicity, speed and repeatability.