| 摘要 |
Disclosed is a method and device for aligning at least two digital images. An embodiment may use frequency-domain transforms of small tiles created from each image to identify substantially similar, distinguishing features within each of the images, and then align the images together based on the location of the distinguishing features. To accomplish this, an embodiment may create equal sized tile sub-images for each image. A key for each tile may be created by performing a frequency-domain transform calculation on each tile. A information-distance difference between each possible pair of tiles on each image may be calculated to identify distinguishing features. From analysis of the information-distance differences of the pairs of tiles, a subset of tiles with high discrimination metrics in relation to other tiles may be located for each image. The subset of distinguishing tiles for each image may then be compared to locate tiles with substantially similar keys and/or information-distance metrics to other tiles of other images. Once similar tiles are located for each image, the images may be aligned in relation to the identified similar tiles. |
