In　order　to　restore　the　original　colours　of　ancient　relics　more　accurately　and　to　reduce　the　burden　of　manual　restoration,　we　developed　a　novel　colour-restoration　technique　based　on　the　DenseNet　algorithm,　which　was　used　in　a　case　study　involving　restoration　of　Dunhuang　mural　images　and　is　based　on　deep　learning.　In　recent　years,　deep　learning-based　methods　have　been　an　important　direction　for　research　into　virtual　restoration　of　image　colours.　Typical,　damaged　murals　were　generated　from　60　mural　datasets　as　input　for　the　system,　and　these　were　enhanced　by　DenseNet,　based　on　the　interactive,　digital　mural-restoration　system.　We　compared　execution　time,　peak　signal-to-noise　ratio　and　structural　similarities　to　evaluate　DenseNet,　SegNet,　Deeplab　and　ResNet　algorithms.　In　terms　of　time　efficiency,　the　DenseNet　algorithm　was　44.62%　faster　than　the　SegNet　algorithm.　Regarding　structural　similarity　(SSIM)　values　for　the　four　models,　DenseNet　was　the　lowest:　1.289%　lower　than　SegNet,　2.442%　lower　than　Deeplab　v3　and　1.288%　lower　than　ResNet.　In　terms　of　the　overall　comparison,　the　repair　effect　for　DenseNet　was　the　best.　Our　method　is　highly　reliable　for　mural　restoration　and　not　only　saves　time　but　also　produces　better　virtual　restoration　results　than　other　methods.