Semantic　object　recognition　is　key　issue　in　CBIR(content　Based　Image　Retrieval),　Video　analysis　and　so　on.　In　this　paper,　we　discuss　how　to　recognize　the　semantic　object　-　basketball.　First,　the　color　feature　is　extracted　to　determine　the　possible　regions　of　basketball,　then　the　steerable　filters　are　used　to　extract　the　MGE　and　MGD,　then　edge　probability　and　the　direction　of　edge　normal　are　obtained　from　MGE　and　MGD.　Next　the　modified　hough　transform　is　designed　to　extract　the　circle　contour.　Finally,　using　template　matching,　the　texture　image　is　extracted　for　object　confirmation.　Experimental　results　confirm　the　efficiency　of　algorithm.