In　this　paper,　a　novel　eyeglasses-face　recognition　approach,　adaptively　doubly　weighted　sub-pattern　LGBP,　is　proposed　to　recognize　eyeglasses-face.　The　proposed　method,　which　looks　forward　to　weak　the　effect　of　eyeglasses　variation,　operates　directly　on　its　sub-patterns　partitioned　from　an　original　whole　pattern　and　separately　extracts　feature　from　them.　Motivated　by　the　fact　that　different　parts　of　the　human　face,　and　convolution　outputs　of　a　sample　image　and　Gabor　kernels,　have　different　contributions,　we　construct　two　weighting　matrices.　Considering　the　instability　of　eyeglasses　as　a　facial　feature,　here　we　make　use　of　3D　face　synthesis　method　based　on　genetic　algorithm　to　reconstruct　virtual　samples,　to　enrich　the　sample　library.　Our　experimental　results　on　FERET　and　Yale　database　reveal　that　the　proposed　approach　is　validity　and　has　better　recognition　performance　than　that　obtained　using　other　traditional　methods.　Copyright　©　2010　Binary　Information　Press.