Photon　mapping　is　an　efficient　method　which　produces　high-quality　photorealistic　images.　Based　on　finite　memory　space,　we　propose　a　new　photon　mapping　algorithm　based　on　noise-reduced　photon　maps　of　external　memory.　Our　method　groups　noise-reduced　photon　maps　corresponding　to　some　original　photon　maps　whose　photons　have　a　lot　of　noise　by　some　stages　of　photon　tracing,　according　to　an　energy　distribution　between　photons.　Then　we　save　these　groups　of　noise-reduced　photon　maps　into　the　external　memory.　In　the　stage　of　illumination　estimation,　we　read　each　group　of　noise-reduced　photon　maps　into　internal　memory　for　producing　the　corresponding　image,　in　order　to　obtain　the　final　image　by　averaging　these　images.　As　these　groups　of　noise-reduced　photon　maps　are　saved　into　external　memory,　we　need　not　to　repeat　a　very　long　stage　of　photon　tracing　as　rendering,　so　the　total　computational　time　of　final　image　is　very　short.　We　can　effectively　reduce　the　bias　of　final　image　by　increasing　the　photon　map　number　of　each　group　of　noise-reduced　photon　maps,　and　reduce　the　noise　of　final　image　by　increasing　the　number　of　original　photon　maps.　Copyright　©　2015　Binary　Information　Press.