The　traditional　3D　face　reconstruction　method　based　on　morphable　model　represents　facial　shape　using　the　knowledge　of　facial　samples.　To　get　good　reconstruction　result,　it　is　necessary　to　use　enough　samples,　such　as　several　hundreds　of　3D　faces.　However,　for　a　given　facial　image,　the　3D　face　reconstruction　is　a　sample　dependency　problem.　Usually　the　similar　samples　will　achieve　good　reconstruction　result,　while　the　dissimilar　samples　will　give　little　benefit　and　even　bring　distortion.　In　this　paper,　we　present　a　3D　face　reconstruction　method　using　merely　the　facial　samples　related　to　the　given　facial　image.　In　the　proposed　method,　the　correlations　between　2D　facial　images　and　3D　faces　are　firstly　established　by　a　training　method　based　on　the　canonical　correlation　analysis　theory.　Then　the　highly　correlated　3D　samples　of　the　given　facial　image　are　selected　according　to　the　correlation　distance.　Based　on　these　selected　samples,　a　compact　3D　morphable　model　is　constructed　for　3D　face　reconstruction.　Matching　the　model　with　the　given　facial　image,　its　3D　face　is　finally　reconstructed.　The　experimental　results　show　that　the　proposed　method　has　good　performance　compared　with　the　traditional　morphable　model.　Copyright　©　2014　Binary　Information　Press.