| 摘要 |
Methods, machines, and machine-readable media for processing multiple exposure source images, including a reference image and one or more non-reference images, are described. In one aspect, respective sets of motion vectors that establish correspondences between respective pairs of the source images are calculated. Ones of the source images are warped to a reference coordinate system based on the sets of motion vectors. For each of the non-reference ones of the warped images, a respective set of one or more measures of alignment confidence between the non-reference warped image and the reference image in the reference coordinate system is computed. Saturated pixels and unsaturated pixels are identified in the warped images. An output image with pixel value contributions from unsaturated pixels in the warped images is generated based on the computed alignment confidence measures.
|
