Realistic Graphics Lab

A Low-Dimensional Function Space for Efficient Spectral Upsampling

To appear in Computer Graphics Forum (Proceedings of Eurographics 2019)

Left. A spec­tral ren­der­ing per­formed us­ing the pro­posed tech­nique. This scene uses a vari­ety of RGB tex­tures that have been con­ver­ted in­to re­flect­ance spec­tra. Right. Plots of high­lighted sur­face re­gions over the vis­ible range.


We present a ver­sat­ile tech­nique to con­vert tex­tures with tri­s­tim­u­lus col­ors in­to the spec­tral do­main, al­low­ing such con­tent to be used in mod­ern ren­der­ing sys­tems. Our meth­od is based on the ob­ser­va­tion that suit­able re­flect­ance spec­tra can be rep­res­en­ted us­ing a low-di­men­sion­al para­met­ric mod­el that is in­trins­ic­ally smooth and en­ergy-con­serving, which leads to sig­ni­fic­ant sim­pli­fic­a­tions com­pared to pri­or work. The res­ult­ing spec­tral tex­tures are com­pact and ef­fi­cient: stor­age re­quire­ments are identic­al to stand­ard RGB tex­tures, and as few as six float­ing point in­struc­tions are re­quired to eval­u­ate them at any wavelength. Our mod­el is the first spec­tral up­sampling meth­od to achieve zero er­ror on the full sR­GB gamut. The tech­nique also sup­ports large-gamut col­or spaces, and can be vec­tor­ized ef­fect­ively for use in ren­der­ing sys­tems that handle many wavelengths at once.

Text citation

Wenzel Jakob and Johannes Hanika. 2019. A Low-Dimensional Function Space for Efficient Spectral Upsampling. In Computer Graphics Forum (Proceedings of Eurographics) 38(2).

    author = {Wenzel Jakob and Johannes Hanika},
    title = {A Low-Dimensional Function Space for Efficient Spectral Upsampling},
    journal = {Computer Graphics Forum (Proceedings of Eurographics)},
    volume = {38},
    number = {2},
    year = {2019},
    month = mar