| 摘要 |
A scene-based non-uniformity correction method employing local constant statistics for eliminating fixed pattern noise in a video comprising a plurality of images is disclosed, comprising the steps of providing an initial gain image, an initial offset image, a predetermined pyramid level, and a filter of a predetermined level; setting one of the plurality of input images to the current image; calculating a temporary true scene image for the current image based on the initial gain image and the initial offset image; accumulating a temporal mean image and a temporal standard deviation image based on the calculated temporary true scene image; setting another of the plurality of images to the current image and repeating the setting, calculating, and accumulating steps until substantially all of the images of the plurality of images have been processed, otherwise further perform the steps of calculating a Gaussian mean image based on the accumulated temporal mean and calculating a Gaussian gain image based on the accumulated temporal standard deviation image; spectrum shaping the Gaussian mean image and the Gaussian gain image based on the predetermined pyramid level and the filter of a predetermined level; multiplying the spectrum shaped Gaussian gain image by the initial gain image to obtain a final gain image; and multiplying the spectrum shaped Gaussian mean image by the initial gain image and add the initial offset image to obtain a final offset image.
|
