In　most　situations,　acquired　digital　images　always　are　corrupted　by　the　wrong　camera　focus,　serious　illumination　even　missing　data.　This　algorithm　is　presented　for　fusion　of　corrupted　multi-sensor　images　by　noise.　Compared　to　the　susceptible　properties　of　PCA　by　large　errors,　the　proposed　method　includes　adaptive　fusion　arithmetic　based　on　matrix　completion　and　self-adaptive　regional　variance　estimation.　Robust　principal　components　analysis　via　inexact　augmented　Lagrange　multiplier　method　is　applied　to　low-frequency　image　components,　and　the　regional　variance　estimation　is　applied　to　high-frequency　components　including　edges　and　details　of　the　original　image.　Compared　with　traditional　algorithms,　the　new　algorithm　not　only　improves　the　amount　of　preserved　information　and　clarity,　but　also　increases　robustness　for　the　corrupted　observation　data.