Efficiently Estimating Projective Transformations

Projective transformations relate the coordinates of images that are taken by either a camera that undergoes only rotation while imaging an arbitrary scene, or one that rotates and translates while imaging a planar surface. Estimating the eight parameters of a projective transformation between a pair of image planes induces a global, dense correspondence between them that can be used for registration or mosaicking. This is a standard problem in image processing and computer vision. The projective transformation estimation problem is typically posed as the minimization of a nonlinear functional of eight parameters, and solved with an "off-the-shelf" numerical algorithm. Here it is shown that in fact, this minimization can be analytically reduced to a nonlinear problem in only two parameters. Any descent algorithm to solve the eight-dimensional minimization can be modified to produce an algorithm for the two-dimensional problem. Several algorithms based on the two-dimensional ...

• Type: Conference
• Year: 2000
• Where: ICIP
• Authors: Richard J. Radke, Peter J. Ramadge, Tomio Echigo, Shun-ichi Iisaku
• Keywords: ICIP 2000, Image Processing, Projective Transformation, Transformation Estimation Problem, Two-dimensional Problem