Cataract　is　one　of　the　leading　causes　of　visual　impairment　worldwide.　People　with　cataracts　often　suffer　a　lot　in　many　aspects　of　daily　life.　Although　early　treatment　can　reduce　the　sufferings　of　cataract　patients　and　prevent　visual　impairment　turning　to　blindness,　people　in　less　developed　areas　still　can＇t　get　timely　treatment　because　of　poor　eye　care　services　or　lack　of　professional　ophthalmologists.　Besides,　the　present　commonly　used　methods　for　cataract　diagnosis,　clinical　assessment　and　photographic　grading,　need　to　be　operated　at　a　slit　lamp　by　ophthalmologists,　which　are　complicated　and　expensive　for　many　patients.　So　reducing　the　cost　and　simplifying　the　process　of　early　cataract　diagnosis　is　of　great　importance.　In　this　paper,　we　proposed　a　fundus　image　based　cataract　classification　method　by　using　pattern　recognition,　which　can　be　used　in　early　screening　of　cataract.　By　calculating　the　2-dimensional　discrete　Fourier　transform　of　a　fundus　image　and　using　the　calculated　spectrum　as　features,　a　cataract　classification　and　grading　method　is　carried　out　by　using　the　linear　discriminant　analysis　promoted　with　the　AdaBoost　algorithm　as　the　classifier.　A　preliminary　test　is　implemented　on　an　image　sample　set　including　460　fundus　images　that　normal,　mild,　moderate　and　severe　cataract　images　are　158,　137,　86　and　79　respectively.　Correspondingly,　the　two-class　and　four-class　classification　accuracy　for　our　proposed　method　are　95.22%　and　81.52%.　We　believe　that　our　proposed　method　has　a　great　potential　in　practical　applications.　©　2014　IEEE.