Many web filters are developed to block porn images from youths. the pron image recognition algorithms can be classified into two kinds: naked skin based and private parts (such as nipples) based. through both of them can keep a lot of porn images away from youths, they would still cause a lot of misrecognitions. this deficiency may not bring obssesions in the daily use of web filters, but it will some other scenarios, it leads to more labors corroborating on pornographic relevant criminal evidences, for example. in this paper, a skin-nipple joint detection proposed to reduce the false positive rate. this joint detection adops gaus mixture model to describe the skin color, and adaboost training algorithm with computer vision related constraints to locate nipples. experiments show that this joint detection approach does not only reach a considerably low false positive rate, but also keep the detection speed at an acceptable level.