This Letter proposes a generalized unified model (GUM) for the calibration of noncentral catadioptric cameras. Releasing the constraint on the projection center and the orientation of the imaging plane that the traditional unified projection model has, the new model is able to well compensate the misalignment between the mirror and the camera. Being a compact and approximate central model, the GUM inherits the flexibility and simplicity from the unified model while maintaining accuracy even under severe misalignment. The calibration algorithm to compute the describing parameters of the model is also given. With the GUM, the calibration of central or noncentral systems could be treated with equal simplicity (or complexity). Experiments on both synthetic data and real images proved our success.Omnidirectional catadioptric cameras featuring the advantage of large field-of-view are being increasingly used in lots of fields such as surveillance and robotics. Calibration is a prerequisite step in most of the applications involving 3D geometries. Depending on whether they pose a single viewpoint, catadioptric cameras can be classified as central or noncentral imaging systems. For central systems, there exist some proper calibration models, among which the unified projection model is the most popular one [1]. Based on this model, a calibration method using planar patterns is proposed by Mei and Rives [2]. The central system has the advantages that the mature computing theories for the perspective cameras are still applicable. However, in practice, most of the catadioptric cameras are noncentral considering the misalignment between the reflective mirror and the camera, let alone the mirror types that do not belong to the limited "central list" such as spherical mirrors.For noncentral systems, great efforts have been made to set up an accurate projection model and compute the describing parameters. Currently there are two main models: the black box model [3] and the complete model [4]. In the black box model, the system is considered a black box and the calibration task is converted into setting up a list of correspondence between the imaging pixels and their corresponding 3D rays. Special structured light patterns [3] could be used to finish this task. Lacking a compact parameterized description, the model suffers low accuracy and flexibility. The complete model [4] considers the system as a combination of a reflective mirror and a projective camera, and models them separately with 20 intrinsic parameters. Computing all of those unknown parameters at one time is difficult due to the highly nonlinear nature of the problem. In [4], a two-step calibration algorithm is proposed by using the black box model [3] first, followed by an iterative bundle adjustment. In particular, assuming the known mirror and the camera's intrinsic parameters, a simplified model concerning only the relative position between the mirror and the camera can be computed by a closed-form solution using the imaged mirror boundary [5]. The largest drawbac...