| 摘要 |
<p>System and method comprising: identifying a patch to be used to fill a target region in an image of a scene, step 702; transforming the patch using depth information of the scene, step 704; filling at least a portion of the destination area using the transformed patch, step 706. Depth data may be calculated from disparity of a pair of stereoscopic images (step 802, fig. 8), or obtained from depth sensors. Transformation may involve scaling, including upscaling and downscaling, that may be uniform or non-uniform. A similarity metric based on texture may be used to select the source patch. Transformation may involve warping using a homography that maps or rotates a three dimensional plane of the source patch to the 3D plane of the fill patch. 3D scene structure may be computed from depth information. Gradients of source regions may be compared to those of fill patches. Scaling may address perspective effects such as foreshortening and may use a ratio of disparities between the patch and target area.</p> |
