Aiming to calibrate the internal parameters of a tilt/shift camera, a flexible self-calibration method with roughly-known calibration pattern is proposed. Based on the imaging characteristics of the tilt/shift camera, two angle parameters are added to explicitly model the homography of the tilted image plane and the ideal image plane which form a part of tilt/shift camera calibration parameter set. Then, a two-step self-calibration algorithm is developed, the first step is to solve the initial parameters of the subset of the calibration parameters via the decomposing of the homography of the imaging plane and the calibration pattern; the second step is a nonlinear optimization technique based on the maximum likelihood criterion. Experimental results confirm the feasibility and validity of the proposed method. The fast convergence of the algorithm shows that the initialization method of the calibration subset parameters is correct and effective. Furthermore, experimental results show that the parameters of the tangential distortion model are coupled with the two-dimensional tilt angle of the image plane. Only the radial distortion model parameters can be taken into account, and the two-dimensional tilt angle with higher accuracy can be obtained.