Capturing a scene with a camera often results in perspective distortion of objects caused by the central projection of the camera. This distortion, although reduced when capturing planar object surfaces from a perpendicular viewpoint, remains inevitable from other angles. In many cases, rectified images, corrected for such distortion, are crucial in diverse fields, including architecture, archaeology, remote sensing, construction, and civil engineering. Traditional photogrammetric methods, such as projective transformation, rely on manually selected control points, which allow to determine the transformation parameters for image rectification. Hence, the automation of this rectification process is essential to scale applications and improve efficiency. In this paper, we present two approaches based on deep learning approaches for the automated perspective rectification of planar object surfaces in images. The first one uses monocular depth estimation methods to initially convert a single-frame image into a 3D point cloud. The RANSAC algorithm is then used to identify the best-fitting plane within the point cloud, such as a building façade in the scene. Subsequently, we orthogonally project the point cloud onto the identified plane, which leads to a perspective-corrected version of the original image. The second approach is inspired by the concept of homography. A regression model based on a deep neural network is presented, which learns the homography parameters to perform the rectification without the need to manually determining any control points. The current results indicate that both approaches are fundamentally viable with the outcomes being moderate, but require further refinement to be practically applicable.