Dynamic Image Stitching with High Exposure Difference

From PanoTools.org Wiki
Revision as of 22:06, 3 April 2009 by Erik Krause (talk | contribs) (categorized)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

Image stitching is a generic approach of generating panorama by combining multiple photographic images with overlapping fields of view and is widely used in applications such as texture synthesis, combining images from multiple angles to form a 3D model and image panorama generation from video. Generating a panorama which could extrapolate motion contained in the input images is difficult and time consuming. In case of armature photography, an introduction to another problem domain arises such as exposure difference resulting in an extremely expensive process. Although lot of work has been done in this domain, unfortunately none of the existing approach deals with all the problems highlighted above simultaneously. Therefore in order to provide a higher degree of accuracy and coverage, we introduce a two step approach which can address all the problems simultaneously.

Step 1: Assuming that the images have been geometrically aligned by some feature detection and matching algorithms such as SIFT and RANSAC, Optimal Cut Evaluation techniques such as Graph Cut can be employed to search for an optimal cut between the input images copied on the output with the help of max flow algorithm. The input images I1 and I2 are then stitched along the optimal cut found in the overlapping region. However the quality of the resultant image R12 will depend upon the level of energy flow through the cut which determines the extent of seam visibility. Graph Cut can thus handle motion upto a certain extent.

Step 2: The second step includes Seam Minimization by using Image Smoothing techniques like seamless editing and cloning with the help of Poisson equation. The core makes use of Poisson partial differential equation with Dirichlet boundary conditions which specifies the Laplacian of an unknown function over the domain of interest along with the unknown function values over the boundary of the domain. We can thus use it to tone a resultant image