Glaucoma is an irreversible blinding eye disease and is difficult to diagnose at an early stage. In this paper, a novel computer-aided diagnosis algorithm was proposed to extract glaucoma features from fundus images and optical coherence tomography (OCT) reports. Based on the ophthalmologists' prior knowledge of the cup-to-disk ratio (CDR) in the fundus images and the thickness of optic nerve fiber layer (RNFLT) in the OCT reports, the proposed algorithm could extract a shared feature of the scale and shape of the target curve (CDR curve or RNFLT curve). This shared feature and features based on prior information constituted the feature space of the target curve, and comprehensively characterized the target curve. Furthermore, based on the feature space of the CDR curve and that of the RNFLT curve, an ensemble classification algorithm was proposed for glaucoma diagnosis. The proposed algorithms were further evaluated on a data set of fundus images and OCT reports. The test results demonstrated that the proposed algorithms achieved the state-of-the-art results with sensitivity of 100%, 97.75% and 91.01%, respectively, at the specificity of 75%, 90% and 99%.