Occlusion Removal with Minimum Number of Images

Published by Institute of Electrical and Electronics Engineers, Inc.

Publication

We examine the problem of combining several images to remove occlusions that occur in one or more of a sequence of images. A sequence of digital camera images of a photoworthy landmark may be occluded by passers-by walking between the photographer and the scene. As they move they will occlude different regions in each image of the sequence. It is obvious that if two or more images coincide at each location an unoccluded image can be formed. We show that this is not necessary: in fact so long as at least one image is unoccluded at each location an unoccluded image can be formed automatically. This greatly eases the conditions under which an unoccluded image can be formed. We detail the algorithm and show results of occlusion removal.