Glaucoma is an irreversible eye disease which is the world's second cause of blindness. The early diagnosis of glaucoma is difficult and computer-aided diagnosis is an important way to solve this problem. Computer-assisted diagnosis of glaucoma faces two major problems: the diagnostic accuracy is not high enough and the principal of the diagnostic algorithm is difficult to be understood by ophthalmologists. In this paper, we designed a new principal component analysis (PCA) algorithm based on prior information and present a quick solution to this algorithm. It is used to extract the features of the cup and disc boundaries from a fundus image for glaucoma diagnosis based on expert knowledge. We obtained the supplemental information for the vertical cup to disc ratio (VCDR) in the scale view of the cup and disc boundaries, and the supplemental information for the global features of the ISNT rule in the shape, view of the boundaries. The information extracted from both views is combined together for glaucoma diagnosis. The algorithm is further validated on a fundus image data set. The test results demonstrated that the proposed algorithm can effectively extract the features of the cup and disc boundaries. In particular, it achieves the state-of-the-art results with a prediction accuracy of more than 95% when combining a variety of classical machine learning classification algorithms. Moreover, the features extracted by the algorithm preserve good interpretability and are easy to be understood and applied by ophthalmologists.