An approach for fully automating perspective images based on symmetry and line intersection

Abstract

Images captured using the camera are perspective in nature and hence has to be converted to a parallel view before being used for any applications to get a photorealistic result. Correction of perspective images are very useful for image based rendering and view metrology. We present a very efficient and simple method for automatically correcting the perspective image based on the intensity variations of objects in the image. The algorithm automates the perspective correction for images and videos (both recorded and online). Here we do not make use of any calibration information or explicit knowledge of any imaging device for carrying out the perspective correction. The method is based on line intersections, Hough Transform and Plane Homography.