Glaucoma is a serious progressive optic neuropathy and is the second leading cause of blindness in the world. In recent years, computer-aided glaucoma diagnosis has gradually attracted more and more attention. While the field has made significant gains, in the existing machine learning algorithm based on fundus images, there is no direct application of the ISNT rule, which is an important criterion for glaucoma diagnosis by doctors. This paper presents a method to quantify the ISNT rule and then extract two features from the optic cup and disk boundary lines. These two features can reflect the cup-to-disc ratio (CDR) and the degree of compliance of the neuroretinal rim to the ISNT rule, respectively. Therefore, a diagnosis classification based on the above features would result in a diagnosis based on doctors' priori knowledge. On a real sample set, the proposed feature extraction and diagnosis algorithms achieve a high prediction accuracy rate.