According　to　the　complexity　of　background,　the　weakness　of　light　and　the　real-time　requirement　of　identification　in　ambient　intelligence　(AmI).　A　face　recognition　mode　based　on　Hidden　Markov　Model　(namely,　HMM)　is　designed,　and　a　complex　feature　extraction　algorithm　based　on　vote　weight　algorithm　which　has　advantages　of　gray　transforms　and　2D-DCT　eigenvectors　is　proposed.　Meanwhile,　the　difference　algorithm　is　applied　to　analyze　each　frame　of　the　picture　captured　by　vidicon,　and　to　position　the　region　of　the　face　in　a　real　time.　At　last,　a　HMM　face　recognition　system　based　on　complex　feature　for　AmI　is　developed.　It　is　demonstrated　experimentally　that　the　system　can　recognize　user＇s　face　well　and　truly,　and　a　base　for　realizing　the　natural　Human-Computer　Interaction　in　ambient　intelligence　is　provided.