Summary of the technology
Project ID : 10-2009-2303
Description of the technology
High quality, fast geometric image editing
- Recent computer vision technologies are enabling geometric image rearrangement.
- Current methods can introduce significant distortions, may limit flexibility or require extensive computation that takes a long time.
- New approach allows objects to be rearranged, removed, and the aspect ratio changed.
- Each operation in the geometric rearrangement of images such as image retargeting, object removal, or object repositioning can be characterised by a shift-map: the relative shift of every pixel in the output image from its source in an input image.
Optimal graph labeling method is used for computing shift-maps; a node in the graph corresponds to a pixel in the output image. In computing the optimal shift-map a data term, which indicates constraints such as the change in image size, object rearrangement, a possible saliency map, etc., and a smoothness term, which minimizes new discontinuities in the output image caused by discontinuities in the shift-map are used.
- Images generated by the shift map are natural looking since distortions that may be introduced by stitching are minimized due to the global smoothness term, the geometric structure of the image is preserved, and large regions can be synthesized.
- Enables global and discrete optimization of an image from which an object has been removed or repositioned
- Very fast and efficient computation; operating on 1M images just take a few seconds
- May be employed for image retargeting, image resizing, image rearrangement such as moving objects or removing them altogether, and image composition from multiple images
Additional Publications by the Researchers
Shmuel Peleg - http://scholar.google.com/citations?user=CshJxRUAAAAJ&hl=en&oi=ao