Standard Non-Uniform Noise Dataset

Fixed Pattern Noise Non-Uniformity Correction through K-Means Clustering

Fixed pattern noise removal from imagery by software correction is a practical approach compared to a physical hardware correction because it allows for correction post-capture of the imagery. Fixed pattern noise presents a unique challenge for de-noising techniques as the noise does not present itself where large number statistics are effective. Traditional noise removal techniques such as blurring or despeckling produce poor correction results because of a lack of noise identification. Other correction methods developed for fixed pattern noise can often present another problem of misidentification of noise. This problem can result in introducing secondary artifacts that can disrupt the imagery and leave the resulting image worse than the uncorrected image. This underlying issue of poor noise identification stems from strong assumptions globally and locally in the imagery. A proposed approach utilizing image intensity clustering will blend local and global information to find a nuanced correction value on a row-by-row basis. The proposed algorithm’s evaluation will be against multiple other correction methods developed for fixed pattern noise removal through a synthetic suite of imagery. The suite is founded on clean images and expanded by varied synthetic noise types introduced by algorithmic means. Images will be evaluated pixel by pixel, row mean by row mean, and with and without a scene intensity bias correction for validation of noise correction.